We introduce HDEdit, a training-free framework for instruction-guided video and 3D scene editing that resolves the fundamental tension between instruction fulfillment and original content preservation through Hierarchical task Decomposition. Our key insight is to progressively decompose complex edits into simpler subtasks. This hierarchical strategy aligns with dual objectives: an LLMguided planner structures high-level subgoals for reliable instruction fulfillment, while embedding-space interpolation further refines each subgoal to preserve unedited content. Two tailored control mechanisms - word-level attention map propagation and parallel denoising synchronization - ensure temporally consistent, hyperparameter tuning-free execution. Beyond video, we extend HDEdit to 3D editing via a simple yet effective render-edit-reconstruct process that maintains strong geometric consistency. Extensive experiments demonstrate our state-of-the-art results across diverse and challenging edits, including longduration videos, fast camera motion, and significant 3D geometric changes.

Achieving realistic simulations of humans interacting with a wide range of objects has long been a fundamental goal. Extending physics-based motion imitation to complex human-object interactions (HOIs) is challenging due to intricate human-object coupling, variability in object geometries, and artifacts in motion capture data, such as inaccurate contacts and limited hand detail. We introduce InterMimic, a framework that enables a single policy to robustly learn from hours of imperfect MoCap data covering diverse full-body interactions with dynamic and varied objects. Our key insight is to employ a curriculum strategy - perfect first, then scale up. We first train subject-specific teacher policies to mimic, retarget, and refine motion capture data. Next, we distill these teachers into a student policy, with the teachers acting as online experts providing direct supervision, as well as high-quality references. Notably, we incorporate RL fine-tuning on the student policy to surpass mere demonstration replication and achieve higher-quality solutions. Our experiments demonstrate that InterMimic produces realistic and diverse interactions across multiple HOI datasets. The learned policy generalizes in a zero-shot manner and seamlessly integrates with kinematic generators, elevating the framework from mere imitation to generative modeling of complex human-object interactions.

While large-scale human motion capture datasets have advanced human motion generation, modeling and generating dynamic 3D human-object interactions (HOIs) remain challenging due to dataset limitations. Existing datasets often lack extensive, high-quality motion and annotation and exhibit artifacts such as contact penetration, floating, and incorrect hand motions. To address these issues, we introduce InterAct, a large-scale 3D HOI benchmark featuring dataset and methodological advancements. First, we consolidate and standardize 21.81 hours of HOI data from diverse sources, enriching it with detailed textual annotations. Second, we propose a unified optimization framework to enhance data quality by reducing artifacts and correcting hand motions. Leveraging the principle of contact invariance, we maintain human-object relationships while introducing motion variations, expanding the dataset to 30.70 hours. Third, we define six benchmarking tasks and develop a unified HOI generative modeling perspective, achieving state-of-the-art performance. Extensive experiments validate the utility of our dataset as a foundational resource for advancing 3D human-object interaction generation. The dataset will be publicly accessible to support further research in the field.

This paper proposes a novel framework utilizing multimodal large language models (MLLMs) for referring video object segmentation (RefVOS). Previous MLLMbased methods commonly struggle with the dilemma between "Ref" and "VOS": they either specialize in understanding a few key frames (global reasoning) or tracking objects on continuous frames (local reasoning), and rely on external VOS or frame selectors to mitigate the other end of the challenge. However, our framework GLUS shows that Global and Local consistency can be Unified into a single video Segmentation MLLM: a set of sparse "context frames" provides global information, while a stream of continuous "query frames" conducts local object tracking. This is further supported by jointly training the MLLM with a pre-trained VOS memory bank to simultaneously digest short-range and long-range temporal information. To improve the information efficiency within the limited context window of MLLMs, we introduce object contrastive learning to distinguish hard false-positive objects and a self-refined framework to identify crucial frames and perform propagation. By collectively integrating these insights, our GLUS delivers a simple yet effective baseline, achieving new state-of-the-art for MLLMs on the MeViS and Ref-Youtube-VOS benchmark. Our project page is at https://glus-video.github.io/.

Recent advances in multimodal large language models (MLLMs) have demonstrated remarkable capabilities in vision-language tasks, yet they often struggle with vision-centric scenarios where precise visual focus is needed for accurate reasoning. In this paper, we introduce Argus to address these limitations with a new visual attention grounding mechanism. Our approach employs object-centric grounding as visual chain-of-thought signals, enabling more effective goal-conditioned visual attention during multimodal reasoning tasks. Evaluations on diverse benchmarks demonstrate that Argus excels in both multi-modal reasoning tasks and referring object grounding tasks. Extensive analysis further validates various design choices of Argus, and reveals the effectiveness of explicit language-guided visual region-of-interest engagement in MLLMs, highlighting the importance of advancing multimodal intelligence from a visual-centric perspective.

Recent advancements in 3D object reconstruction from single images have primarily focused on improving the accuracy of object shapes. Yet, these techniques often fail to accurately capture the inter-relation between the object, ground, and camera. As a result, the reconstructed objects often appear floating or tilted when placed on flat surfaces. This limitation significantly affects 3D-aware image editing applications, like shadow rendering and object pose manipulation. To address this issue, we introduce ORG (Object Reconstruction with Ground), a novel task aimed at reconstructing 3D object geometry in conjunction with the ground surface. Our method uses two compact pixel-level representations to depict the relationship between camera, object, and ground. Experiments show that the proposed ORG model can effectively reconstruct object-ground geometry on unseen data, significantly enhancing the quality of shadow generation and pose manipulation compared to conventional single-image 3D reconstruction techniques. The project page can be found at this website.

We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences - a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5 × acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.

We present REM, a framework for segmenting a wide range of concepts in video that can be described through natural language. Our method leverages the universal visual-language mapping learned by video diffusion models on Internet-scale data by fine-tuning them on small-scale Referring Object Segmentation datasets. Our key insight is to preserve the entirety of the generative model's architecture by shifting its objective from predicting noise to predicting mask latents. The resulting model can accurately segment rare and unseen objects, despite only being trained on a limited set of categories. Additionally, it can effortlessly generalize to non-object dynamic concepts, such as smoke or raindrops, as demonstrated in our new benchmark for Referring Video Process Segmentation (Ref-VPS). REM performs on par with the state-of-the-art on in-domain datasets, like Ref-DAVIS, while outperforming them by up to 12 IoU points out-of-domain, leveraging the power of generative pre-training. We also show that advancements in video generation directly improve segmentation.

Recent image segmentation models have advanced to segment images into high-quality masks for visual entities, and yet they cannot provide comprehensive semantic understanding for complex queries based on both language and vision. This limitation reduces their effectiveness in applications that require user-friendly interactions driven by vision-language prompts. To bridge this gap, we introduce a novel task of omnimodal referring expression segmentation (ORES). In this task, a model produces a group of masks based on arbitrary prompts specified by text only or text plus reference visual entities. To address this new challenge, we propose a novel framework to "Refer to Any Segmentation Mask Group" (RAS), which augments segmentation models with complex multimodal interactions and comprehension via a mask-centric large multimodal model. For training and benchmarking ORES models, we create datasets MaskGroups-2M and MaskGroups-HQ to include diverse mask groups specified by text and reference entities. Through extensive evaluation, we demonstrate superior performance of RAS on our new ORES task, as well as classic referring expression segmentation (RES) and generalized referring expression segmentation (GRES) tasks. Project page: https://Ref2Any.github.io.

The transformative potential of 3D content creation has been progressively unlocked through advancements in generative models. Recently, intuitive drag editing with geometric changes has attracted significant attention in 2D editing yet remains challenging for 3D scenes. In this paper, we introduce 3DGS-Drag, a point-based 3D editing framework that provides efficient, intuitive drag manipulation of real 3D scenes. Our approach bridges the gap between deformation-based and 2D-editing-based 3D editing methods, addressing their limitations to geometry-related content editing. We leverage two key innovations: deformation guidance utilizing 3D Gaussian Splatting for consistent geometric modifications and diffusion guidance for content correction and visual quality enhancement. A progressive editing strategy further supports aggressive 3D drag edits. Our method enables a wide range of edits, including motion change, shape adjustment, inpainting, and content extension. Experimental results demonstrate the effectiveness of 3DGS-Drag in various scenes, achieving state-of-the-art performance in geometry-related 3D content editing. Notably, the editing is efficient, taking 10 to 20 minutes on a single RTX 4090 GPU.

Vision Foundation Models (VFMs) have demonstrated outstanding performance on numerous downstream tasks. However, due to their inherent representation biases originating from different training paradigms, VFMs exhibit advantages and disadvantages across distinct vision tasks. Although amalgamating the strengths of multiple VFMs for downstream tasks is an intuitive strategy, effectively exploiting these biases remains a significant challenge. In this paper, we propose a novel and versatile "Swiss Army Knife" (SAK) solution, which adaptively distills knowledge from a committee of VFMs to enhance multi-task learning. Unlike existing methods that use a single backbone for knowledge transfer, our approach preserves the unique representation bias of each teacher by collaborating the lightweight Teacher-Specific Adapter Path modules with the Teacher-Agnostic Stem. Through dynamic selection and combination of representations with Mixture-of-Representations Routers, our SAK is capable of synergizing the complementary strengths of multiple VFMs. Extensive experiments show that our SAK remarkably outperforms prior state of the arts in multi-task learning by 10% on the NYUD-v2 benchmark, while also providing a flexible and robust framework that can readily accommodate more advanced model designs.

With success in image generation, generative diffusion models are increasingly adopted for discriminative scenarios because generating pixels is a unified and natural perception interface. Although directly re-purposing their generative denoising process has established promising progress in specialist (e.g., depth estimation) and generalist models, the inherent gaps between a generative process and discriminative objectives are rarely investigated. For instance, generative models can tolerate deviations at intermediate sampling steps as long as the final distribution is reasonable, while discriminative tasks with rigorous ground truth for evaluation are sensitive to such errors. Without mitigating such gaps, diffusion for perception still struggles on tasks represented by multi-modal understanding (e.g., referring image segmentation). Motivated by these challenges, we analyze and improve the alignment between the generative diffusion process and perception objectives centering around the key observation: how perception quality evolves with the denoising process. (1) Notably, earlier denoising steps contribute more than later steps, necessitating a tailored learning objective for training: loss functions should reflect varied contributions of timesteps for each perception task. (2) Perception quality drops unexpectedly at later denoising steps, revealing the sensitiveness of perception to training-denoising distribution shift. We introduce diffusion-tailored data augmentation to simulate such drift in the training data. (3) We suggest a novel perspective to the long-standing question: why should a generative process be useful for discriminative tasks - interactivity. The denoising process can be leveraged as a controllable user interface adapting to users' correctional prompts and conducting multi-round interaction in an agentic workflow. Collectively, our insights enhance multiple generative diffusion-based perception models without architectural changes: state-of-the-art diffusion-based depth estimator, previously underplayed referring image segmentation models, and perception generalists. Our code is available at https://github.com/ziqipang/ADDP.

In offline reinforcement learning (RL), managing the distribution shift between the learned policy and the static offline dataset is a persistent challenge that can result in overestimated values and suboptimal policies. Traditional offline RL methods address this by introducing conservative biases that limit exploration to well-understood regions, but they often overly restrict the agent's generalization capabilities. Recent work has sought to generate trajectories using generative models to augment the offline dataset, yet these methods still struggle with overestimating synthesized data, especially when out-of-distribution samples are produced. To overcome this issue, we propose RTDiff, a novel diffusion-based data augmentation technique that synthesizes trajectories *in reverse*, moving from unknown to known states. Such reverse generation naturally mitigates the risk of overestimation by ensuring that the agent avoids planning through unknown states. Additionally, reverse trajectory synthesis allows us to generate longer, more informative trajectories that take full advantage of diffusion models' generative strengths while ensuring reliability. We further enhance RTDiff by introducing flexible trajectory length control and improving the efficiency of the generation process through noise management. Our empirical results show that RTDiff significantly improves the performance of several state-of-the-art offline RL algorithms across diverse environments, achieving consistent and superior results by effectively overcoming distribution shift.

Beyond high-fidelity image synthesis, diffusion models have recently exhibited promising results in dense visual perception tasks. However, most existing work treats diffusion models as a standalone component for perception tasks, employing them either solely for off-the-shelf data augmentation or as mere feature extractors. In contrast to these isolated and thus sub-optimal efforts, we introduce an integrated, versatile, diffusion-based framework, Diff-2-in-1, that can simultaneously handle both multi-modal data generation and dense visual perception, through a unique exploitation of the diffusion-denoising process. Within this framework, we further enhance discriminative visual perception via multi-modal generation, by utilizing the denoising network to create multi-modal data that mirror the distribution of the original training set. Importantly, Diff-2-in-1 optimizes the utilization of the created diverse and faithful data by leveraging a novel self-improving learning mechanism. Comprehensive experimental evaluations validate the effectiveness of our framework, showcasing consistent performance improvements across various discriminative backbones and high-quality multi-modal data generation characterized by both realism and usefulness. Our project website is available at https://zsh2000.github.io/diff-2-in-1.github.io/.

A generalist foundation model agent needs to have a large and diverse skill repertoire, such as finding directions between two travel locations and buying specific items from the Internet. If each skill needs to be specified manually through a fixed set of human-annotated instructions, the agent’s skill repertoire will necessarily be limited due to the scalability of human-annotated instructions. In this work, we address this challenge by proposing Proposer-Agent-Evaluator (PAE), an effective learning system that enables foundation model agents to autonomously discover and practice skills in the wild. After a context-aware task proposer generates instructions based on website information, the agent policy attempts those tasks in the real world with resulting trajectories evaluated by an autonomous VLM-based success evaluator. The success evaluation serves as the reward signal for the agent to refine its policies through RL. We validate PAE on challenging vision-based web navigation, using both real-world and selfhosted websites from WebVoyager and WebArena. Our results show that PAE significantly improves the zero-shot generalization capability of VLM Internet agents (around 50% relative improvement) to both unseen tasks and websites.

This paper proposes Virtual Fitting Room (VFR), a novel video generative model that produces arbitrarily long virtual try-on videos. Our VFR models long video generation tasks as an auto-regressive, segment-by-segment generation process, eliminating the need for resource-intensive generation and lengthy video data, while providing the flexibility to generate videos of arbitrary length. The key challenges of this task are twofold: ensuring local smoothness between adjacent segments and maintaining global temporal consistency across different segments. To address these challenges, we propose our VFR framework, which ensures smoothness through a prefix video condition and enforces consistency with the anchor video — a 360°-view video that comprehensively captures the human's whole-body appearance. Our VFR generates minute-scale virtual try-on videos with both local smoothness and global temporal consistency under various motions, making it a pioneering work in long virtual try-on video generation. Project Page: https://immortalco.github.io/VirtualFittingRoom/.

We present AgMMU, a challenging real‑world benchmark for evaluating and advancing vision-language models (VLMs) in the knowledge‑intensive domain of agriculture. Unlike prior datasets that rely on crowdsourced prompts, AgMMU is distilled from 116,231 authentic dialogues between everyday growers and USDA-authorized Cooperative Extension experts. Through a three‑stage pipeline: automated knowledge extraction, QA generation, and human verification, we construct (i) AgMMU, an evaluation set of 746 multiple‑choice questions (MCQs) and 746 open‑ended questions (OEQs), and (ii) AgBase, a development corpus of 57,079 multimodal facts covering five high-stakes agricultural topics: insect identification, species identification, disease categorization, symptom description, and management instruction. AgMMU has three key advantages:- Authentic & Expert‑Verified: All facts, images, and answers originate from real farmer and gardener inquiries answered by credentialed specialists, ensuring high‑fidelity agricultural knowledge.- Complete Development Suite: AgMMU uniquely couples a dual‑format evaluation benchmark (MCQ and OEQ) with AgBase, a large‑scale training set, enabling both rigorous assessment and targeted improvement of VLMs.- Knowledge‑intensive Challenge: Our tasks demand the synergy of nuanced visual perception and domain expertise, exposing fundamental limitations of current general‑purpose models and charting a path toward robust, application‑ready agricultural AI.Benchmarking 12 leading VLMs reveals pronounced gaps in fine‑grained perception and factual grounding. Open‑sourced models trail after proprietary ones by a wide margin. Simple fine‑tuning on AgBase boosts open-sourced model performance on challenging OEQs for up to 11.6\% on average, narrowing this gap and also motivating future research to propose better strategies in knowledge extraction and distillation from AgBase. We hope AgMMU stimulates research on domain‑specific knowledge integration and trustworthy decision support in agriculture AI development.

The fundamental challenge of long video understanding, e.g., question answering, lies in the extensive number of frames, making it infeasible to densely understand the local details while comprehensively digest the global contexts, especially within a limited context length. To address this problem, our insight is to process short video segments individually and combine these segment-level analyses into a final response. This intuition is noted in the well-established MapReduce principle in big data processing and is naturally compatible with inference scaling at the system level. Motivated by this, we propose MR. Video (pronounced as "mister video"), a long video understanding framework adopting the MapReduce principle. We define the standard operations of MapReduce in a long video understanding context: the Map steps conduct independent and sequence-parallel dense perception on short video segments, covering local details, while the Reduce steps comprehensively aggregate the segment-level results into an answer with global contexts. Thanks to the low cost and convenience of building video agents, we instantiate such Map and Reduce operations as an effective video agent capable of attending to local details and global contexts. Based on such abilities, we further introduce two critical yet previously under-explored long video understanding designs: (a) consistent character/object names in the captions, benefiting the reasoning of actions and stories across long horizons; (b) question intention analysis, which changes the key-frame retrieval in previous video agents to localizing the relevant information via jointly reasoning the whole video contexts and questions. Our MR. Video achieves a >7% accuracy improvement on the challenging LVBench over state-of-the-art video agents and vision-language models (VLMs) and demonstrates a clear advantage on multiple long video benchmarks, highlighting the potential of the MapReduce principle. The code is at https://github.com/ziqipang/MR-Video}{https://github.com/ziqipang/MR-Video.

Foundation models have recently shown strong potential as web agents, capable of interpreting high-level instructions and interacting with complex web interfaces. However, existing training paradigms for these agents often rely on predefined task datasets and curated demonstrations, limiting their scalability, adaptability, and capacity for self-improvement. In this work, we introduce *Self-guided hierArchical exploration for Generalist wEb agents* (SAGE), a new training framework designed to support autonomous skill acquisition through self-guided hierarchical exploration. Our method introduces a three-tier exploration strategy: a pre-exploration phase to build structural understanding of web environments, a top-level exploration strategy to generate a self-evolving curriculum of tasks from easy to hard, and a low-level exploration mechanism that combines planning-based rollouts with step-wise learning to improve policy efficiency. Together, these components form a scalable, supervision-free framework for web agent training. Experimental results on WebVoyager and WebArena demonstrate that our method significantly outperforms prior approaches, enabling foundation model agents to learn complex web tasks with greater generalization and robustness.

Long video understanding is inherently challenging for vision-language models (VLMs) because of the extensive number of frames. With each video frame typically expanding into tens or hundreds of tokens, the limited context length of large language models (LLMs) forces the VLMs to perceive the frames sparsely and lose temporal information. To address this, we explore extreme video token compression towards *one token per frame* at the final LLM layer. Our key insight is that heuristic-based compression, widely adopted by previous methods, is prone to information loss, and this necessitates supervising LLM layers into *learnable* and *progressive* modules for *token-level compression* (LP-Comp). Such compression enables our VLM to digest 2x-4x more frames with improved performance. To further increase the token efficiency, we investigate *frame-level compression*, which selects the frames most relevant to the queries via the internal attention scores of the LLM layers, named *question-conditioned compression* (QC-Comp). As a notable distinction from previous studies, we mitigate the position bias of LLM attention in long contexts, *i.e.*, the over-concentration on the beginning and end of a sequence, by splitting long videos into short segments and employing local attention. Collectively, our combined *token-level* and *frame-level* leads to an e**x**treme compression model for long video understanding, named **XComp**, achieving a significantly larger compression ratio and enabling denser frame sampling. Our XComp is finetuned from VideoChat-Flash with a data-efficient *supervised compression tuning* stage that only requires 2.5\% of the supervised fine-tuning data, yet boosts the accuracy from 42.9\% to 46.2\% on LVBench and enhances multiple other long video benchmarks.

Abstract This paper tackles the limitation of a closed-world object detection model that was trained on one species. The expectation for this model is that it will not generalize well to recognize the instances of new species if they were present in the incoming data stream. We propose a novel object detection framework for this open-world setting that is suitable for applications that monitor wildlife, ocean life, livestock, plant phenotype and crops that typically feature one species in the image. Our method leverages labeled samples from one species in combination with a novelty detection method and Segment Anything Model, a vision foundation model, to (1) identify the presence of new species in unlabeled images, (2) localize their instances, and (3) retrain the initial model with the localized novel class instances. The resulting integrated system assimilates and learns from unlabeled samples of the new classes while not “forgetting” the original species the model was trained on. We demonstrate our findings on two different domains, (1) wildlife detection and (2) plant detection. Our method achieves an AP of 56.2 (for 4 novel species) to 61.6 (for 1 novel species) for wildlife domain, without relying on any ground truth data in the background.

This paper proposes ConsistDreamer - a novel framework that lifts 2D diffusion models with 3D awareness and 3D consistency, thus enabling high-fidelity instruction-guided scene editing. To overcome the fundamental limitation of missing 3D consistency in 2D diffusion models, our key insight is to introduce three synergistic strategies that augment the input of the 2D diffusion model to become 3D-aware and to explicitly enforce 3D consistency during the training process. Specifically, we design sur-rounding views as context-rich input for the 2D diffusion model, and generate 3D-consistent structured noise instead of image-independent noise. Moreover, we introduce self-supervised consistency-enforcing training within the per-scene editing procedure. Extensive evaluation shows that our ConsistDreamer achieves state-of-the-art performance for instruction-guided scene editing across various scenes and editing instructions, particularly in complicated large-scale indoor scenes from ScanNet++, with significantly improved sharpness and fine-grained textures. Notably, ConsistDreamer stands as the first work capable of success-fully editing complex (e.g., plaid/checkered) patterns. Our project page is at immortalco.github.io/ConsistDreamer.

Being able to carry out complicated vision language reasoning tasks in 3D space represents a significant mile-stone in developing household robots and human-centered embodied AI. In this work, we demonstrate that a criti-cal and distinct challenge in 3D vision language reasoning is the situational awareness, which incorporates two key components: (1) The autonomous agent grounds its self-location based on a language prompt. (2) The agent answers open-ended questions from the perspective of its calculated position. To address this challenge, we introduce SIG3D, an end-to-end Situation-Grounded model for 3D vision language reasoning. We tokenize the 3D scene into sparse voxel representation, and propose a language-grounded situation estimator, followed by a situated question answering module. Experiments on the SQA3D and ScanQA datasets show that SIG3D outperforms state-of-the-art models in situational estimation and question answering by a large margin (e.g., an enhancement of over 30% on situation accuracy). Subsequent analysis corrobo-rates our architectural design choices, explores the distinct functions of visual and textual tokens, and highlights the importance of situational awareness in the domain of 3D question-answering. Project page is available at htt ps: //yunzeman.github.io/situation3d.

This paper proposes Instruct 4D-to-4D that achieves 4D awareness and spatial-temporal consistency for 2D diffusion models to generate high-quality instruction-guided dynamic scene editing results. Traditional applications of 2D diffusion models in dynamic scene editing often result in inconsistency, primarily due to their inherent frame-by-frame editing methodology. Addressing the complexities of extending instruction-guided editing to 4D, our key insight is to treat a 4D scene as a pseudo-3D scene, decoupled into two sub-problems: achieving temporal consistency in video editing and applying these edits to the pseudo-3D scene. Following this, we first enhance the Instruct-Pix2Pix (IP2P) model with an anchor-aware attention module for batch processing and consistent editing. Additionally, we integrate optical flow-guided appearance propagation in a sliding window fashion for more precise frame-to-frame editing and incorporate depth-based projection to manage the extensive data of pseudo-3D scenes, followed by iterative editing to achieve convergence. We extensively evaluate our approach in various scenes and editing instructions, and demonstrate that it achieves spatially and temporally consistent editing results, with significantly enhanced detail and sharpness over the prior art. Notably, Instruct 4D-to-4D is general and applicable to both monocular and challenging multi-camera scenes. Code and more results are available at immortalco.github.io/Instruct-4D-to-4D.

We investigate whether region-based representations are effective for recognition. Regions were once a mainstay in recognition approaches, but pixel and patch-based features are now used almost exclusively. We show that recent class-agnostic segmenters like SAM can be effectively combined with strong self-supervised representations, like those from DINOv2, and used for a wide variety of tasks, including semantic segmentation, object-based image re-trieval, and multi-image analysis. Once the masks and features are extracted, these representations, even with linear decoders, enable competitive performance, making them well suited to applications that require custom queries. The representations' compactness also makes them well-suited to video analysis and other problems requiring inference across many images.

The limited scale of current 3D shape datasets hinders the advancements in 3D shape understanding, and motivates multi-modal learning approaches which transfer learned knowledge from data-abundant 2D image and language modalities to 3D shapes. However, even though the image and language representations have been aligned by cross-modal models like CLIP, we find that the image modality fails to contribute as much as the language in existing multi-modal 3D representation learning methods. This is attributed to the domain shift in the 2D images and the distinct focus of each modality. To more effectively leverage both modalities in the pre-training, we introduce TriAdapter Multi-Modal Learning (TAMM) - a novel two-stage learning approach based on three synergistic adapters. First, our CLIP Image Adapter mitigates the domain gap between 3D-rendered images and natural images, by adapting the visual representations of CLIP for synthetic image-text pairs. Subsequently, our Dual Adapters decouple the 3D shape representation space into two complementary sub-spaces: one focusing on visual attributes and the other for semantic understanding, which ensure a more comprehensive and effective multi-modal pre-training. Extensive experiments demonstrate that TAMM consistently enhances 3D representations for a wide range of 3D encoder architectures, pre-training datasets, and downstream tasks. Notably, we boost the zero-shot classification accuracy on Objaverse-LVIS from 46.8% to 50.7%, and improve the 5-way 10-shot linear probing classification accuracy on ModelNet40 from 96.1% to 99.0%. Project page: https://alanzhangcs.github.io/tamm-page.

With recent video object segmentation (VOS) benchmarks evolving to challenging scenarios, we revisit a sim-ple but overlooked strategy: restricting the size of memory banks. This diverges from the prevalent practice of ex-panding memory banks to accommodate extensive histor-ical information. Our specially designed “memory deci-phering” study offers a pivotal insight underpinning such a strategy: expanding memory banks, while seemingly bene-ficial, actually increases the difficulty for VOS modules to decode relevant features due to the confusion from redun-dant information. By restricting memory banks to a limited number of essential frames, we achieve a notable improvement in VOS accuracy. This process balances the im-portance and freshness of frames to maintain an informative memory bank within a bounded capacity. Additionally, restricted memory banks reduce the training-inference discrepancy in memory lengths compared with continuous expansion. This fosters new opportunities in temporal reasoning and enables us to introduce the previously overlooked “temporal positional embedding.” Finally, our insights are embodied in “RMem” (“R” for restricted), a simple yet effective VOS modification that excels at challenging VOS scenarios and establishes new state of the art for object state changes (on the VOST dataset) and long videos (on the Long Videos dataset). Our code and demos are available at https://restricted-memory.github.io/.

Open-world entity segmentation, as an emerging computer vision task, aims at segmenting entities in images without being restricted by pre-defined classes, offering impressive generalization capabilities on unseen images and concepts. Despite its promise, existing entity segmentation methods like Segment Anything Model (SAM) rely heavily on costly expert annotators. This work presents Self-supervised Open-world Hierarchical Entity Segmentation (SOHES), a novel approach that eliminates the need for human annotations. SOHES operates in three phases: self-exploration, self-instruction, and self-correction. Given a pre-trained self-supervised representation, we produce abundant high-quality pseudo-labels through visual feature clustering. Then, we train a segmentation model on the pseudo-labels, and rectify the noises in pseudo-labels via a teacher-student mutual-learning procedure. Beyond segmenting entities, SOHES also captures their constituent parts, providing a hierarchical understanding of visual entities. Using raw images as the sole training data, our method achieves unprecedented performance in self-supervised open-world segmentation, marking a significant milestone towards high-quality open-world entity segmentation in the absence of human-annotated masks. Project page: https://SOHES.github.io.

Protein design, a grand challenge of the day, involves optimization on a fitness landscape, and leading methods adopt a model-based approach where a model is trained on a training set (protein sequences and fitness) and proposes candidates to explore next. These methods are challenged by sparsity of high-fitness samples in the training set, a problem that has been in the literature. A less recognized but equally important problem stems from the distribution of training samples in the design space: leading methods are not designed for scenarios where the desired optimum is in a region that is not only poorly represented in training data, but also relatively far from the highly represented low-fitness regions. We show that this problem of “separation” in the design space is a significant bottleneck in existing model-based optimization tools and propose a new approach that uses a novel VAE as its search model to overcome the problem. We demonstrate its advantage over prior methods in robustly finding improved samples, regardless of the imbalance and separation between low- and high-fitness samples. Our comprehensive benchmark on real and semi-synthetic protein datasets as well as solution design for physics-informed neural networks, showcases the generality of our approach in discrete and continuous design spaces. Our implementation is available at https://github.com/sabagh1994/PGVAE.

This paper reveals that large language models (LLMs), despite being trained solely on text data, are surprisingly}strong encoders for purely visual tasks in the absence of language. Even more intriguingly, this can be achieved by a simple yet previously overlooked strategy -- employing a frozen transformer block from pre-trained LLMs as a constituent encoder layer to directly process visual tokens. Our work pushes the boundaries of leveraging LLMs for computer vision tasks, significantly departing from conventional practices that typically necessitate a multi-modal vision-language setup with associated language prompts, inputs, or outputs. We demonstrate that our approach consistently enhances performance across a diverse range of tasks} encompassing pure 2D or 3D visual recognition tasks (e.g., image and point cloud classification), temporal modeling tasks (e.g., action recognition), non-semantic tasks (e.g., motion forecasting), and multi-modal tasks (e.g., 2D/3D visual question answering and image-text retrieval). Such improvements are a general phenomenon, applicable to various types of LLMs (e.g., LLaMA and OPT) and different LLM transformer blocks. We additionally propose the information filtering hypothesis to explain the effectiveness of pre-trained LLMs in visual encoding -- the pre-trained LLM transformer blocks discern informative visual tokens and further amplify their effect. This hypothesis is empirically supported by the observation that the feature activation, after training with LLM transformer blocks, exhibits a stronger focus on relevant regions. We hope that our work inspires new perspectives on utilizing LLMs and deepening our understanding of their underlying mechanisms.

In real-world scenarios, arbitrary interactions with the environment can often be costly, and actions of expert demonstrations are not always available. To reduce the need for both, offline Learning from Observations (LfO) is extensively studied: the agent learns to solve a task given only expert states and task-agnostic non-expert state-action pairs. The state-of-the-art DIstribution Correction Estimation (DICE) methods, as exemplified by SMODICE, minimize the state occupancy divergence between the learner’s and empirical expert policies. However, such methods are limited to either $f$-divergences (KL and $\chi^2$) or Wasserstein distance with Rubinstein duality, the latter of which constrains the underlying distance metric crucial to the performance of Wasserstein-based solutions. To enable more flexible distance metrics, we propose Primal Wasserstein DICE (PW-DICE). It minimizes the primal Wasserstein distance between the learner and expert state occupancies and leverages a contrastively learned distance metric. Theoretically, our framework is a generalization of SMODICE, and is the first work that unifies $f$-divergence and Wasserstein minimization. Empirically, we find that PW-DICE improves upon several state-of-the-art methods. The code is available at https://github.com/KaiYan289/PW-DICE.

Training autonomous agents with sparse rewards is a long-standing problem in online reinforcement learning (RL), due to low data efficiency. Prior work overcomes this challenge by extracting useful knowledge from offline data, often accomplished through the learning of action distribution from offline data and utilizing the learned distribution to facilitate online RL. However, since the offline data are given and fixed, the extracted knowledge is inherently limited, making it difficult to generalize to new tasks. We propose a novel approach that leverages offline data to learn a generative diffusion model, coined as Adaptive Trajectory Diffuser (ATraDiff). This model generates synthetic trajectories, serving as a form of data augmentation and consequently enhancing the performance of online RL methods. The key strength of our diffuser lies in its adaptability, allowing it to effectively handle varying trajectory lengths and mitigate distribution shifts between online and offline data. Because of its simplicity, ATraDiff seamlessly integrates with a wide spectrum of RL methods. Empirical evaluation shows that ATraDiff consistently achieves state-of-the-art performance across a variety of environments, with particularly pronounced improvements in complicated settings. Our code and demo video are available at https://atradiff.github.io.

While language models (LMs) have shown potential across a range of decision-making tasks, their reliance on simple acting processes limits their broad deployment as autonomous agents. In this paper, we introduce Language Agent Tree Search (LATS) -- the first general framework that synergizes the capabilities of LMs in reasoning, acting, and planning. By leveraging the in-context learning ability of LMs, we integrate Monte Carlo Tree Search into LATS to enable LMs as agents, along with LM-powered value functions and self-reflections for proficient exploration and enhanced decision-making. A key feature of our approach is the incorporation of an environment for external feedback, which offers a more deliberate and adaptive problem-solving mechanism that surpasses the constraints of existing techniques. Our experimental evaluation across diverse domains, including programming, interactive question-answering (QA), web navigation, and math, validates the effectiveness and generality of LATS in decision-making while maintaining competitive or improved reasoning performance. Notably, LATS achieves state-of-the-art pass@1 accuracy (92.7%) for programming on HumanEval with GPT-4 and demonstrates gradient-free performance (average score of 75.9) comparable to gradient-based fine-tuning for web navigation on WebShop with GPT-3.5. Code can be found at https://github.com/lapisrocks/LanguageAgentTreeSearch

Text-conditioned human motion generation has experienced significant advancements with diffusion models trained on extensive motion capture data and corresponding textual annotations. However, extending such success to 3D dynamic human-object interaction (HOI) generation faces notable challenges, primarily due to the lack of large-scale interaction data and comprehensive descriptions that align with these interactions. This paper takes the initiative and showcases the potential of generating human-object interactions without direct training on text-interaction pair data. Our key insight in achieving this is that interaction semantics and dynamics can be decoupled. Being unable to learn interaction semantics through supervised training, we instead leverage pre-trained large models, synergizing knowledge from a large language model and a text-to-motion model. While such knowledge offers high-level control over interaction semantics, it cannot grasp the intricacies of low-level interaction dynamics. To overcome this issue, we introduce a world model designed to comprehend simple physics, modeling how human actions influence object motion. By integrating these components, our novel framework, InterDreamer, is able to generate text-aligned 3D HOI sequences without relying on paired text-interaction data. We apply InterDreamer to the BEHAVE, OMOMO, and CHAIRS datasets, and our comprehensive experimental analysis demonstrates its capability to generate realistic and coherent interaction sequences that seamlessly align with the text directives.

This paper proposes ProEdit - a simple yet effective framework for high-quality 3D scene editing guided by diffusion distillation in a novel progressive manner. Inspired by the crucial observation that multi-view inconsistency in scene editing is rooted in the diffusion model’s large feasible output space (FOS), our framework controls the size of FOS and reduces inconsistency by decomposing the overall editing task into several subtasks, which are then executed progressively on the scene. Within this framework, we design a difficulty-aware subtask decomposition scheduler and an adaptive 3D Gaussian splatting (3DGS) training strategy, ensuring high efficiency in performing each subtask. Extensive evaluation shows that our ProEdit achieves state-of-the-art results in various scenes and challenging editing tasks, all through a simple framework without any expensive or sophisticated add-ons like distillation losses, components, or training procedures. Notably, ProEdit also provides a new way to preview, control, and select the aggressivity of editing operation during the editing process.

In this paper, we approach an overlooked yet critical task Graph2Image: generating images from multimodal attributed graphs (MMAGs). This task poses significant challenges due to the explosion in graph size, dependencies among graph entities, and the need for controllability in graph conditions. To address these challenges, we propose a graph context-conditioned diffusion model called InstructG2I. InstructG2I first exploits the graph structure and multimodal information to conduct informative neighbor sampling by combining personalized page rank and re-ranking based on vision-language features. Then, a graph QFormer encoder adaptively encodes the graph nodes into an auxiliary set of graph prompts to guide the denoising process of diffusion. Finally, we propose graph classifier-free guidance, enabling controllable generation by varying the strength of graph guidance and multiple connected edges to a node. Extensive experiments conducted on three datasets from different domains demonstrate the effectiveness and controllability of our approach. The code is available at https://github.com/PeterGriffinJin/InstructG2I.

Complex 3D scene understanding has gained increasing attention, with scene encoding strategies built on top of visual foundation models playing a crucial role in this success. However, the optimal scene encoding strategies for various scenarios remain unclear, particularly compared to their image-based counterparts. To address this issue, we present the first comprehensive study that probes various visual encoding models for 3D scene understanding, identifying the strengths and limitations of each model across different scenarios. Our evaluation spans seven vision foundation encoders, including image, video, and 3D foundation models. We evaluate these models in four tasks: Vision-Language Scene Reasoning, Visual Grounding, Segmentation, and Registration, each focusing on different aspects of scene understanding. Our evaluation yields key intriguing findings: Unsupervised image foundation models demonstrate superior overall performance, video models excel in object-level tasks, diffusion models benefit geometric tasks, language-pretrained models show unexpected limitations in language-related tasks, and the mixture-of-vision-expert (MoVE) strategy leads to consistent performance improvement. These insights challenge some conventional understandings, provide novel perspectives on leveraging visual foundation models, and highlight the need for more flexible encoder selection in future vision-language and scene understanding tasks.

Decision Transformers have recently emerged as a new and compelling paradigm for offline Reinforcement Learning (RL), completing a trajectory in an autoregressive way. While improvements have been made to overcome initial shortcomings, online finetuning of decision transformers has been surprisingly under-explored. The widely adopted state-of-the-art Online Decision Transformer (ODT) still struggles when pretrained with low-reward offline data. In this paper, we theoretically analyze the online-finetuning of the decision transformer, showing that the commonly used Return-To-Go (RTG) that's far from the expected return hampers the online fine-tuning process. This problem, however, is well-addressed by the value function and advantage of standard RL algorithms. As suggested by our analysis, in our experiments, we hence find that simply adding TD3 gradients to the finetuning process of ODT effectively improves the online finetuning performance of ODT, especially if ODT is pretrained with low-reward offline data. These findings provide new directions to further improve decision transformers.

The creation of complex 3D scenes tailored to user specifications has been a tedious and challenging task with traditional 3D modeling tools. Although some pioneering methods have achieved automatic text-to-3D generation, they are generally limited to small-scale scenes with restricted control over the shape and texture. We introduce SceneCraft, a novel method for generating detailed indoor scenes that adhere to textual descriptions and spatial layout preferences provided by users. Central to our method is a rendering-based technique, which converts 3D semantic layouts into multi-view 2D proxy maps. Furthermore, we design a semantic and depth conditioned diffusion model to generate multi-view images, which are used to learn a neural radiance field (NeRF) as the final scene representation. Without the constraints of panorama image generation, we surpass previous methods in supporting complicated indoor space generation beyond a single room, even as complicated as a whole multi-bedroom apartment with irregular shapes and layouts. Through experimental analysis, we demonstrate that our method significantly outperforms existing approaches in complex indoor scene generation with diverse textures, consistent geometry, and realistic visual quality.

Object discovery – separating objects from the background without manual labels – is a fundamental open challenge in computer vision. Previous methods struggle to go beyond clustering of low-level cues, whether handcrafted (e.g., color, texture) or learned (e.g., from auto-encoders). In this work, we augment the auto-encoder representation learning framework with two key components: motion-guidance and mid-level feature tokenization. Although both have been separately investigated, we introduce a new transformer decoder showing that their benefits can compound thanks to motion-guided vector quantization. We show that our architecture effectively leverages the synergy between motion and tokenization, improving upon the state of the art on both synthetic and real datasets. Our approach enables the emergence of interpretable object-specific mid-level features, demonstrating the benefits of motion-guidance (no labeling) and quantization (interpretability, memory efficiency).

Object detectors often suffer from the domain gap between training (source domain) and real-world applications (target domain). Mean-teacher self-training is a powerful paradigm in unsupervised domain adaptation for object detection, but it struggles with low-quality pseudo-labels. In this work, we identify the intriguing alignment and synergy between mean-teacher self-training and contrastive learning. Motivated by this, we propose Contrastive Mean Teacher (CMT) - a unified, general-purpose framework with the two paradigms naturally integrated to maximize beneficial learning signals. Instead of using pseudo-labels solely for final predictions, our strategy extracts object-level features using pseudo-labels and optimizes them via contrastive learning, without requiring labels in the target domain. When combined with recent mean-teacher self-training methods, CMT leads to new state-of-the-art target-domain performance: 51.9% mAP on Foggy Cityscapes, outperforming the previously best by 2.1% mAP. Notably, CMT can stabilize performance and provide more significant gains as pseudo-label noise increases.

This paper proposes NeuralEditor that enables neural radiance fields (NeRFs) natively editable for general shape editing tasks. Despite their impressive results on novel-view synthesis, it remains a fundamental challenge for NeRFs to edit the shape of the scene. Our key insight is to exploit the explicit point cloud representation as the underlying structure to construct NeRFs, inspired by the intuitive interpretation of NeRF rendering as a process that projects or “plots” the associated 3D point cloud to a 2D image plane. To this end, NeuralEditor introduces a novel rendering scheme based on deterministic integration within K-D tree-guided density-adaptive voxels, which produces both high-quality rendering results and precise point clouds through optimization. NeuralEditor then performs shape editing via mapping associated points between point clouds. Extensive evaluation shows that NeuralEditor achieves state-of-the-art performance in both shape deformation and scene morphing tasks. Notably, NeuralEditor supports both zeroshot inference and further fine-tuning over the edited scene. Our code, benchmark, and demo video are available at immortalco.github.io/NeuralEditor.

Integrating multiple sensors and addressing diverse tasks in an end-to-end algorithm are challenging yet critical topics for autonomous driving. To this end, we introduce BEVGuide, a novel Bird's Eye- View (BEV) representation learning framework, representing the first attempt to unify a wide range of sensors under direct BEV guidance in an end-to-end fashion. Our architecture accepts input from a diverse sensor pool, including but not limited to Camera, Lidar and Radar sensors, and extracts BEV feature embeddings using a versatile and general transformer backbone. We design a BEV-guided multi-sensor attention block to take queries from BEV embeddings and learn the BEV representation from sensor-specific features. BEVGuide is efficient due to its lightweight backbone design and highly flexible as it supports almost any input sensor configurations. Extensive experiments demonstrate that our framework achieves exceptional performance in BEV perception tasks with a diverse sensor set. Project page is at https://yunzeman.github.io/BEVGuide.

This work proposes an end-to-end multi-camera 3D multi-object tracking (MOT) framework. It emphasizes spatio-temporal continuity and integrates both past and future reasoning for tracked objects. Thus, we name it “Past- and-Future reasoning for Tracking” (PF-Track). Specifically, our method adopts the “tracking by attention” framework and represents tracked instances coherently over time with object queries. To explicitly use historical cues, our “Past Reasoning” module learns to refine the tracks and enhance the object features by cross-attending to queries from previous frames and other objects. The “Future Reasoning” module digests historical information and predicts robust future trajectories. In the case of long-term occlusions, our method maintains the object positions and enables re-association by integrating motion predictions. On the nuScenes dataset, our method improves AMOTA by a large margin and remarkably reduces ID-Switches by 90% compared to prior approaches, which is an order of magnitude less. The code and models are made available at https://github.com/TRI-ML/PF-Track.

This paper addresses a novel task of anticipating 3D human-object interactions (HOIs). Most existing research on HOI synthesis lacks comprehensive whole-body interactions with dynamic objects, e.g., often limited to manipulating small or static objects. Our task is significantly more challenging, as it requires modeling dynamic objects with various shapes, capturing whole-body motion, and ensuring physically valid interactions. To this end, we propose InterDiff, a framework comprising two key steps: (i) interaction diffusion, where we leverage a diffusion model to encode the distribution of future human-object interactions; (ii) interaction correction, where we introduce a physics-informed predictor to correct denoised HOIs in a diffusion step. Our key insight is to inject prior knowledge that the interactions under reference with respect to contact points follow a simple pattern and are easily predictable. Experiments on multiple human-object interaction datasets demonstrate the effectiveness of our method for this task, capable of producing realistic, vivid, and remarkably longterm 3D HOI predictions.

While bird’s-eye-view (BEV) perception models can be helpful in building high-definition maps (HD maps) with less human labor, their results are often unreliable and demonstrate noticeable inconsistencies in the predicted HD maps from different viewpoints. This is because BEV perception is typically set up in an "onboard" manner, which restricts the computation and prevents algorithms from simultaneously reasoning multiple views. This paper overcomes these limitations and advocates a more practical "offboard" HD map generation setup that removes the computation constraints, based on the fact that HD maps are commonly reusable infrastructures built offline in data centers. To this end, we propose a novel offboard pipeline called MV-Map that capitalizes multi-view consistency and can handle an arbitrary number of frames with the key design of a "region-centric" framework. In MV-Map, the target HD maps are created by aggregating all the frames of onboard predictions, weighted by the confidence scores assigned by an "uncertainty network." To further enhance multi-view consistency, we augment the uncertainty network with the global 3D structure optimized by a voxelized neural radiance field (Voxel-NeRF). Extensive experiments on nuScenes show that our MV-Map significantly improves the quality of HD maps, further highlighting the importance of offboard methods for HD map generation. Our code and model are available at https://github.com/ZiYang-xie/MV-Map.

Daily objects commonly experience state changes. For example, slicing a cucumber changes its state from whole to sliced. Learning about object state changes in Video Object Segmentation (VOS) is crucial for understanding and interacting with the visual world. Conventional VOS benchmarks do not consider this challenging yet crucial problem. This paper makes a pioneering effort to introduce a weakly-supervised benchmark on Video State-Changing Object Segmentation (VSCOS). We construct our VSCOS benchmark by selecting state-changing videos from existing datasets. In advocate of an annotation-efficient approach towards state-changing object segmentation, we only annotate the first and last frames of training videos, which is different from conventional VOS. Notably, an open-vocabulary setting is included to evaluate the generalization to novel types of objects or state changes. We empirically illustrate that state-of-the-art VOS models struggle with state-changing objects and lose track after the state changes. We analyze the main difficulties of our VSCOS task and identify three technical improvements, namely, fine-tuning strategies, representation learning, and integrating motion information. Applying these improvements results in a strong baseline for segmenting state-changing objects consistently. Our benchmark and baseline methods are publicly available at https://github.com/venom12138/VSCOS.

Multi-task visual learning is a critical aspect of computer vision. Current research, however, predominantly concentrates on the multi-task dense prediction setting, which overlooks the intrinsic 3D world and its multi-view consistent structures, and lacks the capability for versatile imagination. In response to these limitations, we present a novel problem setting – multi-task view synthesis (MTVS), which reinterprets multi-task prediction as a set of novel-view synthesis tasks for multiple scene properties, including RGB. To tackle the MTVS problem, we propose MuvieNeRF, a framework that incorporates both multi-task and cross-view knowledge to simultaneously synthesize multiple scene properties. MuvieNeRF integrates two key modules, the Cross-Task Attention (CTA) and Cross-View Attention (CVA) modules, enabling the efficient use of information across multiple views and tasks. Extensive evaluation on both synthetic and realistic benchmarks demonstrates that MuvieNeRF is capable of simultaneously synthesizing different scene properties with promising visual quality, even outperforming conventional discriminative models in various settings. Notably, we show that MuvieNeRF exhibits universal applicability across a range of NeRF backbones. Our code is available at https://github.com/zsh2000/MuvieNeRF.

Dense depth and surface normal predictors should possess the equivariant property to cropping-and-resizing – cropping the input image should result in cropping the same output image. However, we find that state-of-the-art depth and normal predictors, despite having strong performances, surprisingly do not respect equivariance. The problem exists even when crop-and-resize data augmentation is employed during training. To remedy this, we propose an equivariant regularization technique, consisting of an averaging procedure and a self-consistency loss, to explicitly promote cropping-and-resizing equivariance in depth and normal networks. Our approach can be applied to both CNN and Transformer architectures, does not incur extra cost during testing, and notably improves the supervised and semi-supervised learning performance of dense predictors on Taskonomy tasks. Finally, finetuning with our loss on unlabeled images improves not only equivariance but also accuracy of state-of-the-art depth and normal predictors when evaluated on NYU-v2.

Though self-supervised contrastive learning (CL) has shown its potential to achieve state-of-the-art accuracy without any supervision, its behavior still remains under-investigated. Different from most previous work that understands CL from learning objectives, we focus on an unexplored yet natural aspect: the spatial inductive bias which seems to be implicitly exploited via data augmentations in CL. We design an experiment to study the reliance of CL on such spatial inductive bias, by destroying the global or local spatial structures of an image with global or local patch shuffling, and comparing the performance drop between experiments on original and corrupted dataset to quantify the reliance on certain inductive bias. We also use the uniformity of feature space to further research how CL-pre-trained models behave with the corrupted dataset. Our results and analysis show that CL has a much higher reliance on spatial inductive bias than SL, regardless of specific CL algorithm or backbones, opening a new direction for studying the behavior of CL.

Resource-constrained perception systems such as edge computing and vision-for-robotics require vision models to be both accurate and lightweight in computation and memory usage. While knowledge distillation is a proven strategy to enhance the performance of lightweight classification models, its application to structured outputs like object detection and instance segmentation remains a complicated task, due to the variability in outputs and complex internal network modules involved in the distillation process. In this paper, we propose a simple yet surprisingly effective sequential approach to knowledge distillation that progressively transfers the knowledge of a set of teacher detectors to a given lightweight student. To distill knowledge from a highly accurate but complex teacher model, we construct a sequence of teachers to help the student gradually adapt. Our progressive strategy can be easily combined with existing detection distillation mechanisms to consistently maximize student performance in various settings. To the best of our knowledge, we are the first to successfully distill knowledge from Transformer-based teacher detectors to convolution-based students, and unprecedentedly boost the performance of ResNet-50 based RetinaNet from 36.5% to 42.0% AP and Mask R-CNN from 38.2% to 42.5% AP on the MS COCO benchmark. Code available at https://github.com/Shengcao-Cao/MTPD.

Closing the domain gap between training and deployment and incorporating multiple sensor modalities are two challenging yet critical topics for self-driving. Existing work only focuses on single one of the above topics, overlooking the simultaneous domain and modality shift which pervasively exists in real-world scenarios. A model trained with multi-sensor data collected in Europe may need to run in Asia with a subset of input sensors available. In this work, we propose DualCross, a cross-modality cross-domain adaptation framework to facilitate the learning of a more robust monocular bird's-eye-view (BEV) perception model, which transfers the point cloud knowledge from a LiDAR sensor in one domain during the training phase to the camera-only testing scenario in a different domain. This work results in the first open analysis of cross-domain cross-sensor perception and adaptation for monocular 3D tasks in the wild. We benchmark our approach on large-scale datasets under a wide range of domain shifts and show state-of-the-art results against various baselines. Our project webpage is at https://yunzeman.github.io/DualCross.

Trajectory forecasting is a widely-studied problem for autonomous navigation. However, existing benchmarks evaluate forecasting based on independent snapshots of trajectories, which are not representative of real-world applications that operate on a continuous stream of data. To bridge this gap, we introduce a benchmark that continuously queries future trajectories on streaming data and we refer to it as “streaming forecasting.” Our benchmark inherently captures the disappearance and re-appearance of agents, presenting the emergent challenge of forecasting for occluded agents, which is a safetycritical problem yet overlooked by snapshot-based benchmarks. Moreover, forecasting in the context of continuous timestamps naturally asks for temporal coherence between predictions from adjacent timestamps. Based on this benchmark, we further provide solutions and analysis for streaming forecasting. We propose a plug-and-play meta-algorithm called “Predictive Streamer” that can adapt any snapshot-based forecaster into a streaming forecaster. Our algorithm estimates the states of occluded agents by propagating their positions with multi-modal trajectories, and leverages differentiable filters to ensure temporal consistency. Both occlusion reasoning and temporal coherence strategies significantly improve forecasting quality, resulting in 25% smaller endpoint errors for occluded agents and 10-20% smaller fluctuations of trajectories. Our work is intended to generate interest within the community by highlighting the importance of addressing motion forecasting in its intrinsic streaming setting. Code is available at https://github.com/ziqipang/StreamingForecasting.

Depth completion, which aims to generate high-quality dense depth maps from sparse depth maps, has attracted increasing attention in recent years. Previous work usually employs RGB images as guidance, and introduces iterative spatial propagation to refine estimated coarse depth maps. However, most of the propagation refinement methods require several iterations and suffer from a fixed receptive field, which may contain irrelevant and useless information with very sparse input. In this paper, we address these two challenges simultaneously by revisiting the idea of deformable convolution. We propose an effective architecture that leverages deformable kernel convolution as a single-pass refinement module, and empirically demonstrate its superiority. To better understand the function of deformable convolution and exploit it for depth completion, we further systematically investigate a variety of representative strategies. Our study reveals that, different from prior work, deformable convolution needs to be applied on an estimated depth map with a relatively high density for better performance. We evaluate our model on the large-scale KITTI dataset and achieve state-of-the-art level performance in both accuracy and inference speed. Our code is available at https://github.com/AlexSunNiklReDC.

The human visual perception system demonstrates exceptional capabilities in learning without explicit supervision and understanding the part-to-whole composition of objects. Drawing inspiration from these two abilities, we propose Hierarchical Adaptive Self-Supervised Object Detection (HASSOD), a novel approach that learns to detect objects and understand their compositions without human supervision. HASSOD employs a hierarchical adaptive clustering strategy to group regions into object masks based on self-supervised visual representations, adaptively determining the number of objects per image. Furthermore, HASSOD identifies the hierarchical levels of objects in terms of composition, by analyzing coverage relations between masks and constructing tree structures. This additional self-supervised learning task leads to improved detection performance and enhanced interpretability. Lastly, we abandon the inefficient multi-round self-training process utilized in prior methods and instead adapt the Mean Teacher framework from semi-supervised learning, which leads to a smoother and more efficient training process. Through extensive experiments on prevalent image datasets, we demonstrate the superiority of HASSOD over existing methods, thereby advancing the state of the art in self-supervised object detection. Notably, we improve Mask AR from 20.2 to 22.5 on LVIS, and from 17.0 to 26.0 on SA-1B. Project page: https://HASSOD-NeurIPS23.github.io.

We introduce ViCA-NeRF, the *first* view-consistency-aware method for 3D editing with text instructions. In addition to the implicit neural radiance field (NeRF) modeling, our key insight is to exploit two sources of regularization that *explicitly* propagate the editing information across different views, thus ensuring multi-view consistency. For *geometric regularization*, we leverage the depth information derived from NeRF to establish image correspondences between different views. For *learned regularization*, we align the latent codes in the 2D diffusion model between edited and unedited images, enabling us to edit key views and propagate the update throughout the entire scene. Incorporating these two strategies, our ViCA-NeRF operates in two stages. In the initial stage, we blend edits from different views to create a preliminary 3D edit. This is followed by a second stage of NeRF training, dedicated to further refining the scene's appearance. Experimental results demonstrate that ViCA-NeRF provides more flexible, efficient (3 times faster) editing with higher levels of consistency and details, compared with the state of the art. Our code is available at: https://github.com/Dongjiahua/VICA-NeRF

Offline imitation from observations aims to solve MDPs where only task-specific expert states and task-agnostic non-expert state-action pairs are available. Offline imitation is useful in real-world scenarios where arbitrary interactions are costly and expert actions are unavailable. The state-of-the-art ‘DIstribution Correction Estimation’ (DICE) methods minimize divergence of state occupancy between expert and learner policies and retrieve a policy with weighted behavior cloning; however, their results are unstable when learning from incomplete trajectories, due to a non-robust optimization in the dual domain. To address the issue, in this paper, we propose Trajectory-Aware Imitation Learning from Observations (TAILO). TAILO uses a discounted sum along the future trajectory as the weight for weighted behavior cloning. The terms for the sum are scaled by the output of a discriminator, which aims to identify expert states. Despite simplicity, TAILO works well if there exist trajectories or segments of expert behavior in the task-agnostic data, a common assumption in prior work. In experiments across multiple testbeds, we find TAILO to be more robust and effective, particularly with incomplete trajectories.

We propose a conceptually simple and lightweight framework for improving the robustness of vision models through the combination of knowledge distillation and data augmentation. We address the conjecture that larger models do not make for better teachers by showing strong gains in out-of-distribution robustness when distilling from pretrained foundation models. Following this finding, we propose Discrete Adversarial Distillation (DAD), which leverages a robust teacher to generate adversarial examples and a VQGAN to discretize them, creating more informative samples than standard data augmentation techniques. We provide a theoretical framework for the use of a robust teacher in the knowledge distillation with data augmentation setting and demonstrate strong gains in out-of-distribution robustness and clean accuracy across different student architectures. Notably, our method adds minor computational overhead compared to similar techniques and can be easily combined with other data augmentations for further improvements.

The healthcare and AI communities have witnessed a growing interest in the development of AI-assisted systems for automated diagnosis of Parkinson's Disease (PD), one of the most prevalent neurodegenerative disorders. However, the progress in this area has been significantly impeded by the absence of a unified, publicly available benchmark, which prevents comprehensive evaluation of existing PD analysis methods and the development of advanced models. This work overcomes these challenges by introducing YouTubePD -- the first publicly available multimodal benchmark designed for PD analysis. We crowd-source existing videos featured with PD from YouTube, exploit multimodal information including in-the-wild videos, audio data, and facial landmarks across 200+ subject videos, and provide dense and diverse annotations from clinical expert. Based on our benchmark, we propose three challenging and complementary tasks encompassing both discriminative and generative tasks, along with a comprehensive set of corresponding baselines. Experimental evaluation showcases the potential of modern deep learning and computer vision techniques, in particular the generalizability of the models developed on YouTubePD to real-world clinical settings, while revealing their limitations. We hope our work paves the way for future research in this direction.

Existing work on continual learning (CL) is primarily devoted to developing algorithms for models trained from scratch. Despite their encouraging performance on contrived benchmarks, these algorithms show dramatic performance drop in real-world scenarios. Therefore, this paper advocates the systematic introduction of pre-training to CL, which is a general recipe for transferring knowledge to downstream tasks but is substantially missing in the CL community. Our investigation reveals the multifaceted complexity of exploiting pre-trained models for CL, along three different axes: pre-trained models, CL algorithms, and CL scenarios. Perhaps most intriguingly, improvements in CL algorithms from pre-training are very inconsistent – an underperforming algorithm could become competitive and even state of the art, when all algorithms start from a pretrained model. This indicates that the current paradigm, where all CL methods are compared in from-scratch training, is not well reflective of the true CL objective and desired progress. In addition, we make several other important observations, including that 1) CL algorithms that exert less regularization benefit more from a pre-trained model; and 2) a stronger pre-trained model such as CLIP does not guarantee a better improvement. Based on these findings, we introduce a simple yet effective baseline that employs minimum regularization and leverages the more beneficial pre-trained model, coupled with a two-stage training pipeline. We recommend including this strong baseline in the future development of CL algorithms, due to its demonstrated state-of-the-art performance. Our code is available at https://github.com/eric11220/pretrained-models-in-CL.

Comprehensive 3D scene understanding, both geometrically and semantically, is important for real-world applications such as robot perception. Most of the existing work has focused on developing data-driven discriminative models for scene understanding. This paper provides a new approach to scene understanding, from a synthesis model perspective, by leveraging the recent progress on implicit scene representation and neural rendering. Building upon the great success of Neural Radiance Fields (NeRFs), we introduce Scene-Property Synthesis with NeRF (SS-NeRF) that is able to not only render photo-realistic RGB images from novel viewpoints, but also render various accurate scene properties (e.g., appearance, geometry, and semantics). By doing so, we facilitate addressing a variety of scene understanding tasks under a unified framework, including semantic segmentation, surface normal estimation, reshading, keypoint detection, and edge detection. Our SS-NeRF framework can be a powerful tool for bridging generative learning and discriminative learning, and thus be beneficial to the investigation of a wide range of interesting problems, such as studying task relationships within a synthesis paradigm, transferring knowledge to novel tasks, facilitating downstream discriminative tasks as ways of data augmentation, and serving as auto-labeller for data creation. Our code is available at https://github.com/zsh2000/SS-NeRF.

In the real open world, data tends to follow long-tailed class distributions, motivating the well-studied long-tailed recognition (LTR) problem. Naive training produces models that are biased toward common classes in terms of higher accuracy. The key to addressing LTR is to balance various aspects including data distribution, training losses, and gradients in learning. We explore an orthogonal direction, weight balancing, motivated by the empirical observation that the naively trained classifier has “artificially” larger weights in norm for common classes (because there exists abundant data to train them, unlike the rare classes). We investigate three techniques to balance weights, L2-normalization, weight decay, and MaxNorm. We first point out that L2-normalization “perfectly” balances per-class weights to be unit norm, but such a hard constraint might prevent classes from learning better classifiers. In contrast, weight decay penalizes larger weights more heavily and so learns small balanced weights; the MaxNorm constraint encourages growing small weights within a norm ball but caps all the weights by the radius. Our extensive study shows that both help learn balanced weights and greatly improve the LTR accuracy. Surprisingly, weight decay, although underexplored in LTR, significantly improves over prior work. Therefore, we adopt a two-stage training paradigm and propose a simple approach to LTR: (1) learning features using the cross-entropy loss by tuning weight decay, and (2) learning classifiers using class-balanced loss by tuning weight decay and MaxNorm. Our approach achieves the state-of-the-art accuracy on five standard benchmarks, serving as a future baseline for long-tailed recognition.

This paper studies the problem of object discovery - separating objects from the background without manual labels. Existing approaches utilize appearance cues, such as color, texture, and location, to group pixels into object-like regions. However, by relying on appearance alone, these methods fail to separate objects from the background in cluttered scenes. This is a fundamental limitation since the definition of an object is inherently ambiguous and context-dependent. To resolve this ambiguity, we choose to focus on dynamic objects - entities that can move independently in the world. We then scale the recent auto-encoder based frameworks for unsuper-vised object discovery from toy synthetic images to complex real-world scenes. To this end, we simplify their architecture, and augment the resulting model with a weak learning signal from general motion segmentation algorithms. Our experiments demonstrate that, despite only capturing a small subset of the objects that move, this signal is enough to generalize to segment both moving and static instances of dynamic objects. We show that our model scales to a newly collected, photo- realistic synthetic dataset with street driving scenarios. Additionally, we leverage ground truth segmentation and flow annotations in this dataset for thorough ablation and evaluation. Finally, our experiments on the real-world KITTI benchmark demonstrate that the proposed approach outperforms both heuristic- and learning-based methods by capitalizing on motion cues.

In LiDAR-based 3D object detection for autonomous driving, the ratio of the object size to input scene size is significantly smaller compared to 2D detection cases. Over-looking this difference, many 3D detectors directly follow the common practice of 2D detectors, which downsample the feature maps even after quantizing the point clouds. In this paper, we start by rethinking how such multi-stride stereotype affects the LiDAR-based 3D object detectors. Our experiments point out that the downsampling operations bring few advantages, and lead to inevitable information loss. To remedy this issue, we propose Single-stride Sparse Transformer (SST) to maintain the original resolution from the beginning to the end of the network. Armed with transformers, our method addresses the problem of insufficient receptive field in single-stride architectures. It also cooperates well with the sparsity of point clouds and naturally avoids expensive computation. Eventually, our SST achieves state-of-the-art results on the large-scale Waymo Open Dataset. It is worth mentioning that our method can achieve exciting performance (83.8 LEVEL_1 AP on validation split) on small object (pedestrian) detection due to the characteristic of single stride. Our codes will be public soon.

DIVeR builds on the key ideas of NeRF and its variants-density models and volume rendering – to learn 3D object models that can be rendered realistically from small numbers of images. In contrast to all previous NeRF methods, DIVeR uses deterministic rather than stochastic estimates of the volume rendering integral. DIVeR's representation is a voxel based field of features. To compute the volume rendering integral, a ray is broken into intervals, one per voxel; components of the volume rendering integral are estimated from the features for each interval using an MLP, and the components are aggregated. As a result, DIVeR can render thin translucent structures that are missed by other integrators. Furthermore, DIVeR's representation has semantics that is relatively exposed compared to other such methods – moving feature vectors around in the voxel space results in natural edits. Extensive qualitative and quantitative comparisons to current state-of-the-art methods show that DIVeR produces models that (1) render at or above state-of-the-art quality, (2) are very small without being baked, (3) render very fast without being baked, and (4) can be edited in natural ways. Our real-time code is available at: https://github.com/lwwu2/diver-rt

For real-world graph data, the node class distribution is inherently imbalanced and long-tailed, which naturally leads to a few-shot learning scenario with limited nodes labeled for newly emerging classes. Existing efforts are carefully designed to solve such a few-shot learning problem via data augmentation, learning transferable initialization, to name a few. However, most, if not all, of them are based on a strong assumption that all the test nodes must exclusively come from novel classes, which is impractical in real-world applications. In this paper, we study a broader and more realistic problem named generalized few-shot node classification, where the test samples can be from both novel classes and base classes. Compared with the standard fewshot node classification, this new problem imposes several unique challenges, including asymmetric classification and inconsistent preference. To counter those challenges, we propose a shot-aware graph neural network (STAGER) equipped with an uncertainty-based weight assigner module for adaptive propagation. To formulate this problem from the meta-learning perspective, we propose a new training paradigm named imbalanced episodic training to ensure the label distribution is consistent between the training and test scenarios. Experiment results on four real-world datasets demonstrate the efficacy of our model, with up to 14% accuracy improvement over baselines.

Generative modeling has recently shown great promise in computer vision, but it has mostly focused on synthesizing visually realistic images. In this paper, motivated by multi-task learning of shareable feature representations, we consider a novel problem of learning a shared generative model that is useful across various visual perception tasks. Correspondingly, we propose a general multi-task oriented generative modeling (MGM) framework, by coupling a discriminative multi-task network with a generative network. While it is challenging to synthesize both RGB images and pixel-level annotations in multi-task scenarios, our framework enables us to use synthesized images paired with only weak annotations (i.e., image-level scene labels) to facilitate multiple visual tasks. Experimental evaluation on challenging multi-task benchmarks, including NYUv2 and Taskonomy, demonstrates that our MGM framework improves the performance of all the tasks by large margins, consistently outperforming state-of-the-art multi-task approaches in different sample-size regimes.

Lifelong learners must recognize concept vocabularies that evolve over time. A common yet underexplored scenario is learning with class labels that continually refine/expand old classes. For example, humans learn to recognize ${\tt dog}$ before dog breeds. In practical settings, dataset ${\it versioning}$ often introduces refinement to ontologies, such as autonomous vehicle benchmarks that refine a previous ${\tt vehicle}$ class into ${\tt school-bus}$ as autonomous operations expand to new cities. This paper formalizes a protocol for studying the problem of ${\it Learning with Evolving Class Ontology}$ (LECO). LECO requires learning classifiers in distinct time periods (TPs); each TP introduces a new ontology of "fine" labels that refines old ontologies of "coarse" labels (e.g., dog breeds that refine the previous ${\tt dog}$). LECO explores such questions as whether to annotate new data or relabel the old, how to exploit coarse labels, and whether to finetune the previous TP's model or train from scratch. To answer these questions, we leverage insights from related problems such as class-incremental learning. We validate them under the LECO protocol through the lens of image classification (on CIFAR and iNaturalist) and semantic segmentation (on Mapillary). Extensive experiments lead to some surprising conclusions; while the current status quo in the field is to relabel existing datasets with new class ontologies (such as COCO-to-LVIS or Mapillary1.2-to-2.0), LECO demonstrates that a far better strategy is to annotate ${\it new}$ data with the new ontology. However, this produces an aggregate dataset with inconsistent old-vs-new labels, complicating learning. To address this challenge, we adopt methods from semi-supervised and partial-label learning. We demonstrate that such strategies can surprisingly be made near-optimal, in the sense of approaching an "oracle" that learns on the aggregate dataset exhaustively labeled with the newest ontology.

Although reinforcement learning has found widespread use in dense reward settings, training autonomous agents with sparse rewards remains challenging. To address this difficulty, prior work has shown promising results when using not only task-specific demonstrations but also task-agnostic albeit somewhat related demonstrations. In most cases, the available demonstrations are distilled into an implicit prior, commonly represented via a single deep net. Explicit priors in the form of a database that can be queried have also been shown to lead to encouraging results. To better benefit from available demonstrations, we develop a method to Combine Explicit and Implicit Priors (CEIP). CEIP exploits multiple implicit priors in the form of normalizing flows in parallel to form a single complex prior. Moreover, CEIP uses an effective explicit retrieval and push-forward mechanism to condition the implicit priors. In three challenging environments, we find the proposed CEIP method to improve upon sophisticated state-of-the-art techniques.

Learning to detect novel objects from few annotated examples is of great practical importance. A particularly challenging yet common regime occurs when there are extremely limited examples (less than three). One critical factor in improving few-shot detection is to address the lack of variation in training data. We propose to build a better model of variation for novel classes by transferring the shared within-class variation from base classes. To this end, we introduce a hallucinator network that learns to generate additional, useful training examples in the region of interest (RoI) feature space, and incorporate it into a modern object detection model. Our approach yields significant performance improvements on two state-of-the-art few-shot detectors with different proposal generation procedures. In particular, we achieve new state of the art in the extremely-few-shot regime on the challenging COCO benchmark.

This paper presents a detection-aware pre-training (DAP) approach, which leverages only weakly-labeled classification-style datasets (e.g., ImageNet) for pretraining, but is specifically tailored to benefit object detection tasks. In contrast to the widely used image classification-based pre-training (e.g., on ImageNet), which does not include any location-related training tasks, we transform a classification dataset into a detection dataset through a weakly supervised object localization method based on Class Activation Maps to directly pre-train a detector, making the pre-trained model location-aware and capable of predicting bounding boxes. We show that DAP can outperform the traditional classification pre-training in terms of both sample efficiency and convergence speed in downstream detection tasks including VOC and COCO. In particular, DAP boosts the detection accuracy by a large margin when the number of examples in the downstream task is small.

Few-shot classification aims at classifying categories of a novel task by learning from just a few (typically, 1 to 5) labelled examples. An effective approach to few-shot classification involves a prior model trained on a large-sample base domain, which is then finetuned over the novel few-shot task to yield generalizable representations. However, task-specific finetuning is prone to overfitting due to the lack of enough training examples. To alleviate this issue, we propose a new finetuning approach based on contrastive learning that reuses unlabelled examples from the base do-main in the form of distractors. Unlike the nature of unlabelled data used in prior works, distractors belong to classes that do not overlap with the novel categories. We demonstrate for the first time that inclusion of such distractors can significantly boost few-shot generalization. Our technical novelty includes a stochastic pairing of examples sharing the same category in the few-shot task and a weighting term that controls the relative influence of task-specific negatives and distractors. An important aspect of our finetuning objective is that it is agnostic to distractor labels and hence applicable to various base domain settings. More precisely, compared to state-of-the-art approaches, our method shows accuracy gains of up to 12% in cross-domain and up to 5% in unsupervised prior-learning settings. Our code is available at https://github.com/quantacode/Contrastive-Finetuning.git

Learning to hallucinate additional examples has recently been shown as a promising direction to address few-shot learning tasks. This work investigates two important yet overlooked natural supervision signals for guiding the hallucination process – (i) extrinsic: classifiers trained on hallucinated examples should be close to strong classifiers that would be learned from a large amount of real examples; and (ii) intrinsic: clusters of hallucinated and real examples belonging to the same class should be pulled together, while simultaneously pushing apart clusters of hallucinated and real examples from different classes. We achieve (i) by introducing an additional mentor model on data-abundant base classes for directing the hallucinator, and achieve (ii) by performing contrastive learning between hallucinated and real examples. As a general, model-agnostic framework, our dual mentor-and self-directed (DMAS) hallucinator significantly improves few-shot learning performance on widely-used benchmarks in various scenarios.

We present a novel semi-supervised semantic segmentation method which jointly achieves two desiderata of segmentation model regularities: the label-space consistency property between image augmentations and the feature-space contrastive property among different pixels. We lever-age the pixel-level ℓ2loss and the pixel contrastive loss for the two purposes respectively. To address the computational efficiency issue and the false negative noise issue involved in the pixel contrastive loss, we further introduce and investigate several negative sampling techniques. Extensive experiments demonstrate the state-of-the-art performance of our method (PC2Seg) with the DeepLab-v3+ architecture, in several challenging semi-supervised settings derived from the VOC, Cityscapes, and COCO datasets.

Training on datasets with long-tailed distributions has been challenging for major recognition tasks such as classification and detection. To deal with this challenge, image resampling is typically introduced as a simple but effective approach. However, we observe that long-tailed detection differs from classification since multiple classes may be present in one image. As a result, image resampling alone is not enough to yield a sufficiently balanced distribution at the object-level. We address object-level resampling by introducing an object-centric sampling strategy based on a dynamic, episodic memory bank. Our proposed strategy has two benefits: 1) convenient object-level resampling without significant extra computation, and 2) implicit feature-level augmentation from model updates. We show that image-level and object-level resamplings are both important, and thus unify them with a joint resampling strategy. Our method achieves state-of-the-art performance on the rare categories of LVIS, with 1.89% and 3.13% relative improvements over Forest R-CNN on detection and instance segmentation.

Humans can robustly learn novel visual concepts even when images undergo various deformations and loose certain information. Mimicking the same behavior and synthesizing deformed instances of new concepts may help visual recognition systems perform better one-shot learning, i.e., learning concepts from one or few examples. Our key insight is that, while the deformed images may not be visually realistic, they still maintain critical semantic information and contribute significantly to formulating classifier decision boundaries. Inspired by the recent progress of meta-learning, we combine a meta-learner with an image deformation sub-network that produces additional training examples, and optimize both models in an end-to-end manner. The deformation sub-network learns to deform images by fusing a pair of images --- a probe image that keeps the visual content and a gallery image that diversifies the deformations. We demonstrate results on the widely used one-shot learning benchmarks (miniImageNet and ImageNet 1K Challenge datasets), which significantly outperform state-of-the-art approaches.

One of the key limitations of modern deep learning approaches lies in the amount of data required to train them. Humans, by contrast, can learn to recognize novel categories from just a few examples. Instrumental to this rapid learning ability is the compositional structure of concept representations in the human brain --- something that deep learning models are lacking. In this work, we make a step towards bridging this gap between human and machine learning by introducing a simple regularization technique that allows the learned representation to be decomposable into parts. Our method uses category-level attribute annotations to disentangle the feature space of a network into subspaces corresponding to the attributes. These attributes can be either purely visual, like object parts, or more abstract, like openness and symmetry. We demonstrate the value of compositional representations on three datasets: CUB-200-2011, SUN397, and ImageNet, and show that they require fewer examples to learn classifiers for novel categories.

Few-shot learning, i.e., learning novel concepts from few examples, is fundamental to practical visual recognition systems. While most of existing work has focused on few-shot classification, we make a step towards few-shot object detection, a more challenging yet under-explored task. We develop a conceptually simple but powerful meta-learning based framework that simultaneously tackles few-shot classification and few-shot localization in a unified, coherent way. This framework leverages meta-level knowledge about "model parameter generation" from base classes with abundant data to facilitate the generation of a detector for novel classes. Our key insight is to disentangle the learning of category-agnostic and category-specific components in a CNN based detection model. In particular, we introduce a weight prediction meta-model that enables predicting the parameters of category-specific components from few examples. We systematically benchmark the performance of modern detectors in the small-sample size regime. Experiments in a variety of realistic scenarios, including within-domain, cross-domain, and long-tailed settings, demonstrate the effectiveness and generality of our approach under different notions of novel classes.

One-shot learning aims to recognize novel target classes from few examples by transferring knowledge from source classes, under a general assumption that the source and target classes are semantically related but not exactly the same. Based on this assumption, recent work has focused on image-based one-shot learning, while little work has addressed video-based one shot learning. One of the challenges lies in that it is difficult to maintain the disjoint-class assumption for videos, since video clips of target classes may potentially appear in the videos of source classes. To address this issue, we introduce a novel setting, termed as embodied agents based one-shot learning, which leverages synthetic videos produced in a virtual environment to understand realistic videos of target classes. In this setting, we further propose two types of learning tasks: embodied one-shot video domain adaptation and embodied one-shot video transfer recognition. These tasks serve as a testbed for evaluating video related one-shot learning tasks. In addition, we propose a general video segment augmentation method, which significantly facilitates a variety of one-shot learning tasks. Experimental results validate the soundness of our setting and learning tasks, and also show the effectiveness of our augmentation approach to video recognition in the small-sample size regime.

Humans can quickly learn new visual concepts, perhaps because they can easily visualize or imagine what novel objects look like from different views. Incorporating this ability to hallucinate novel instances of new concepts might help machine vision systems perform better low-shot learning, i.e., learning concepts from few examples. We present a novel approach to low-shot learning that uses this idea. Our approach builds on recent progress in meta-learning ("learning to learn") by combining a meta-learner with a "hallucinator" that produces additional training examples, and optimizing both models jointly. Our hallucinator can be incorporated into a variety of meta-learners and provides significant gains: up to a 6 point boost in classification accuracy when only a single training example is available, yielding state-of-the-art performance on the challenging ImageNet low-shot classification benchmark.

Teaching a robot to predict and mimic how a human moves or acts in the near future by observing a series of historical human movements is a crucial first step in human-robot interaction and collaboration. In this paper, we instrument a robot with such a prediction ability by leveraging recent deep learning and computer vision techniques. First, our system takes images from the robot camera as input to produce the corresponding human skeleton based on real-time human pose estimation obtained with the OpenPose library. Then, conditioning on this historical sequence, the robot forecasts plausible motion through a motion predictor, generating a corresponding demonstration. Because of a lack of high-level fidelity validation, existing forecasting algorithms suffer from error accumulation and inaccurate prediction. Inspired by generative adversarial networks (GANs), we introduce a global discriminator that examines whether the predicted sequence is smooth and realistic. Our resulting motion GAN model achieves superior prediction performance to state-of-the-art approaches when evaluated on the standard H3.6M dataset. Based on this motion GAN model, the robot demonstrates its ability to replay the predicted motion in a human-like manner when interacting with a person.

Deep convolutional neural networks (CNNs) have recognized promise as universal representations for various image recognition tasks. One of their properties is the ability to transfer knowledge from a large annotated source dataset (e.g., ImageNet) to a (typically smaller) target dataset. This is usually accomplished through supervised fine-tuning on labeled new target data. In this work, we address "unsupervised fine-tuning" that transfers a pre-trained network to target tasks with unlabeled data such as image clustering tasks. To this end, we introduce group-sparse non-negative matrix factorization (GSNMF), a variant of NMF, to identify a rich set of high-level latent variables that are informative on the target task. The resulting "factorized convolutional network" (FCN) can itself be seen as a feed-forward model that combines CNN and two-layer structured NMF. We empirically validate our approach and demonstrate state-of-the-art image clustering performance on challenging scene (MIT-67) and fine-grained (Birds-200, Flowers-102) benchmarks. We further show that, when used as unsupervised initialization, our approach improves image classification performance as well.

CNNs have made an undeniable impact on computer vision through the ability to learn high-capacity models with large annotated training sets. One of their remarkable properties is the ability to transfer knowledge from a large source dataset to a (typically smaller) target dataset. This is usually accomplished through fine-tuning a fixed-size network on new target data. Indeed, virtually every contemporary visual recognition system makes use of fine-tuning to transfer knowledge from ImageNet. In this work, we analyze what components and parameters change during fine-tuning, and discover that increasing model capacity allows for more natural model adaptation through fine-tuning. By making an analogy to developmental learning, we demonstrate that growing a CNN with additional units, either by widening existing layers or deepening the overall network, significantly outperforms classic fine-tuning approaches. But in order to properly grow a network, we show that newly-added units must be appropriately normalized to allow for a pace of learning that is consistent with existing units. We empirically validate our approach on several benchmark datasets, producing state-of-the-art results.

We describe an approach to learning from long-tailed, imbalanced datasets that are prevalent in real-world settings. Here, the challenge is to learn accurate "few-shot'' models for classes in the tail of the class distribution, for which little data is available. We cast this problem as transfer learning, where knowledge from the data-rich classes in the head of the distribution is transferred to the data-poor classes in the tail. Our key insights are as follows. First, we propose to transfer meta-knowledge about learning-to-learn from the head classes. This knowledge is encoded with a meta-network that operates on the space of model parameters, that is trained to predict many-shot model parameters from few-shot model parameters. Second, we transfer this meta-knowledge in a progressive manner, from classes in the head to the "body'', and from the "body'' to the tail. That is, we transfer knowledge in a gradual fashion, regularizing meta-networks for few-shot regression with those trained with more training data. This allows our final network to capture a notion of model dynamics, that predicts how model parameters are likely to change as more training data is gradually added. We demonstrate results on image classification datasets (SUN, Places, and ImageNet) tuned for the long-tailed setting, that significantly outperform common heuristics, such as data resampling or reweighting.

Category classifiers trained from a large corpus of annotated data are widely accepted as the sources for (hypothesis) transfer learning. Sources generated in this way are tied to a particular set of categories, limiting their transferability across a wide spectrum of target categories. In this paper, we address this largely-overlooked yet fundamental source problem by both introducing a systematic scheme for generating universal source hypotheses and proposing a principled, scalable approach to automatically tuning the transfer process. Our approach is based on the insights that expressive source hypotheses could be generated without any supervision and that a sparse combination of such hypotheses facilitates recognition of novel categories from few samples. We demonstrate improvements over the state-of-the-art on object and scene classification in the small sample size regime.

This work explores CNNs for the recognition of novel categories from few examples. Inspired by the transferability properties of CNNs, we introduce an additional unsupervised meta-training stage that exposes multiple top layer units to a large amount of unlabeled real-world images. By encouraging these units to learn diverse sets of low-density separators across the unlabeled data, we capture a more generic, richer description of the visual world, which decouples these units from ties to a specific set of categories. We propose an unsupervised margin maximization that jointly estimates compact high-density regions and infers low-density separators. The low-density separator (LDS) modules can be plugged into any or all of the top layers of a standard CNN architecture. The resulting CNNs significantly improve the performance in scene classification, fine-grained recognition, and action recognition with small training samples.

In this paper, we explore an approach to generating detectors that is radically different from the conventional way of learning a detector from a large corpus of annotated positive and negative data samples. Instead, we assume that we have evaluated “off-line” a large library of detectors against a large set of detection tasks. Given a new target task, we evaluate a subset of the models on few samples from the new task and we use the matrix of models-tasks ratings to predict the performance of all the models in the library on the new task, enabling us to select a good set of detectors for the new task. This approach has three key advantages of great interest in practice: 1) generating a large collection of expressive models in an unsupervised manner is possible; 2) a far smaller set of annotated samples is needed compared to that required for training from scratch; and 3) recommending models is a very fast operation compared to the notoriously expensive training procedures of modern detectors. (1) will make the models informative across different categories; (2) will dramatically reduce the need for manually annotating vast datasets for training detectors; and (3) will enable rapid generation of new detectors.

The current sparse representation framework is to decouple it as two subproblems, i.e., alternate sparse coding and dictionary learning using different optimizers, treating elements in bases and codes separately. In this paper, we treat elements both in bases and codes ho-mogenously. The original optimization is directly decoupled as several blockwise alternate subproblems rather than above two. Hence, sparse coding and bases learning optimizations are coupled together. And the variables involved in the optimization problems are partitioned into several suitable blocks with convexity preserved, making it possible to perform an exact block coordinate descent. For each separable subproblem, based on the convexity and monotonic property of the parabolic function, a closed-form solution is obtained. Thus the algorithm is simple, efficient and effective. Experimental results show that our algorithm significantly accelerates the learning process.

Collaborative representation based classification (CRC) has been successfully used for visual recognition and showed impressive performance recently. However, it directly uses the training samples from each class as the subspaces to calculate the minimum residual error for a given testing sample. This leads to high residual error and instability, which is critical especially for a small number of training samples in each class. In this paper, we propose a class specific subspace learning algorithm for collaborative representation. By introducing the dual form of subspace learning, it presents an explicit relationship between the basis vectors and the original image features, and thus enhances the interpretability. Lagrange multipliers are then applied to optimize the corresponding objective function, i.e., learning the weights used in constructing the subspaces. Extensive experimental results demonstrate that the proposed algorithm has achieved superior performance in several visual recognition tasks.

Sparse representation technique has been widely used in various areas of computer vision over the last decades. Unfortunately, in the current formulations, there are no explicit relationship between the learned dictionary and the original data. By tracing back and connecting sparse representation with the K-means algorithm, a novel variation scheme termed as self-explanatory convex sparse representation (SCSR) has been proposed in this paper. To be specific, the basis vectors of the dictionary are refined as convex combination of the data points. The atoms now would capture a notion of centroids similar to K-means, leading to enhanced interpretability. Sparse representation and K-means are thus unified under the same framework in this sense. Besides, an appealing property also emerges that the weight and code matrices both tend to be naturally sparse without additional constraints. Compared with the standard formulations, SCSR is easier to be extended into the kernel space. To solve the corresponding sparse coding sub problem and dictionary learning sub problem, block-wise coordinate descent and Lagrange multipliers are proposed accordingly. To validate the proposed algorithm, it is implemented in image classification, a successful applications of sparse representation. Experimental results on several benchmark data sets, such as UIUC-Sports, Scene 15, and Caltech-256 demonstrate the effectiveness of our proposed algorithm.

Nonnegative Matrix Factorization (NMF), a relatively novel paradigm for dimensionality reduction, has been in the ascendant since its inception. It incorporates the nonnegativity constraint and thus obtains the parts-based representation as well as enhancing the interpretability of the issue correspondingly. This survey paper mainly focuses on the theoretical research into NMF over the last 5 years, where the principles, basic models, properties, and algorithms of NMF along with its various modifications, extensions, and generalizations are summarized systematically. The existing NMF algorithms are divided into four categories: Basic NMF (BNMF), Constrained NMF (CNMF), Structured NMF (SNMF), and Generalized NMF (GNMF), upon which the design principles, characteristics, problems, relationships, and evolution of these algorithms are presented and analyzed comprehensively. Some related work not on NMF that NMF should learn from or has connections with is involved too. Moreover, some open issues remained to be solved are discussed. Several relevant application areas of NMF are also briefly described. This survey aims to construct an integrated, state-of-the-art framework for NMF concept, from which the follow-up research may benefit.

Recently, dictionary learned by sparse coding has been widely adopted in image classification and has achieved competitive performance. Sparse coding is capable of reducing the reconstruction error in transforming low-level descriptors into compact mid-level features. Nevertheless, dictionary learned by sparse coding does not have the ability to distinguish different classes. That is to say, it is not the optimum dictionary for the classification task. In this paper, based on the global image statistics, a novel discriminant dictionary learning method combining linear discriminant analysis with sparse coding is proposed to obtain a more discriminative dictionary while preserving its descriptive abilities and a block coordinate descent algorithm is proposed to solve the optimization problem. Experimental results show that our algorithm has capabilities to learn dictionary with more discriminative power and achieves superior performance.

Non-negative Matrix Factorization (NMF) has become a powerful tool for image representation due to its enhanced semantic interpretability under non-negativity. Unfortunately, two types of neighborhood information essential to representation are lost in NMF. For individual image, the local structure information is missing in the vectorization, which can then be avoided by Non-negative Tensor Factorization (NTF). For image data points, they often reside on a low dimensional submanifold embedded in a high dimensional ambient space. NMF and NTF are incapable of encoding the local geometrical information, which can nevertheless be resuscitated by manifold learning. To simultaneously model both of the neighborhood relationship within and among image data, this paper proposes a novel algorithm called Neighborhood Preserving Non-negative Tensor Factorization (NPNTF) by incorporating locally linear embedding regularization into tensor factorization. Experimental results on image clustering show the superior performance of NPNTF with more natural and discriminating representation ability.

This paper proposes a novel patch propagation inpainting algorithm based on Weighted Sparse Non-negative Matrix Factorization (WSNMF). Unlike existing methods, we cast the inpainting task as a sequential low-rank matrix recovery and completion problem, where the incomplete data matrix consists of the image patch to be inpainted and several similar intact candidate patches under the assumption that they can be described using a low-dimensional linear model. Besides, the non-negativity and sparsity constraints are enforced for the additive sparse linear combination. The WSNMF, based on the Expectation-Maximization (EM) procedure, is then introduced to predict missing values. Experimental results show that this approach exploits the available information from the source region more adequately and thus has capabilities to recover both structure and composite textures more effectively as well as preventing unwanted artifacts compared to current exemplar-based techniques.