Autoregressive models, built based on the Next Token Prediction (NTP) paradigm, show great potential in developing a unified framework that integrates both language and vision tasks. Pioneering works introduce NTP to autoregressive visual generation tasks. In this work, we rethink the NTP for autoregressive image generation and extend it to a novel Next Patch Prediction (NPP) paradigm. Our key idea is to group and aggregate image tokens into patch tokens with higher information density. By using patch tokens as a more compact input sequence, the autoregressive model is trained to predict the next patch, significantly reducing computational costs. To further exploit the natural hierarchical structure of image data, we propose a multi-scale coarse-to-fine patch grouping strategy. With this strategy, the training process begins with a large patch size and ends with vanilla NTP where the patch size is 1x1, thus maintaining the original inference process without modifications. Extensive experiments across a diverse range of model sizes demonstrate that NPP could reduce the training cost to around 0.6 times while improving image generation quality by up to 1.0 FID score on the ImageNet 256x256 generation benchmark. Notably, our method retains the original autoregressive model architecture without introducing additional trainable parameters or specifically designing a custom image tokenizer, offering a flexible and plug-and-play solution for enhancing autoregressive visual generation.

Recent advances insingle-view3D scene reconstruction have highlighted the challenges in capturing fine geometric details and ensuring structural consistency, particularly in high-fidelity outdoor scene modeling. This paper presents Niagara, a new single-view 3D scene reconstruction framework that can faithfully reconstruct challenging outdoor scenes from a single input image for the first time. Our approach integrates monocular depth and normal estimation as input, which substantially improves its ability to capture fine details, mitigating common issues like geometric detail loss and deformation. Additionally, we introduce a geometric affine field (GAF) and 3D self-attention as geometry-constraint, which combines the structural properties of explicit geometry with the adaptability of implicit feature fields, striking a balance between efficient rendering and high-fidelity reconstruction. Our framework finally proposes a specialized encoder-decoder architecture, where a depth-based 3D Gaussian decoder is proposed to predict 3D Gaussian parameters, which can be used for novel view synthesis. Extensive results and analyses suggest that our Niagara surpasses prior SoTA approaches such as Flash3D in both single-view and dual-view settings, significantly enhancing the geometric accuracy and visual fidelity, especially in outdoor scenes. Webpage: https://ai-kunkun.github.io/Niagara page/.

Multimodal Large Language Models (MLLMs) have displayed remarkable performance in multi-modal tasks, particularly in visual comprehension. However, we reveal that MLLMs often generate incorrect answers even when they understand the visual content. To this end, we manually construct a benchmark with 12 categories and design evaluation metrics that assess the degree of error in MLLM responses even when the visual content is seemingly understood. Based on this benchmark, we test 15 leading MLLMs and analyze the distribution of attention maps and logits of some MLLMs. Our investigation identifies two primary issues: 1) most instruction tuning datasets predominantly feature questions that "directly" relate to the visual content, leading to a bias in MLLMs’ responses to other indirect questions, and 2) MLLMs’ attention to visual tokens is notably lower than to system and question tokens. We further observe that attention scores between questions and visual tokens as well as the model’s confidence in the answers are lower in response to misleading questions than to straightforward ones. To address the first challenge, we introduce a paired positive and negative data construction pipeline to diversify the dataset. For the second challenge, we propose to enhance the model’s focus on visual content during decoding by refining the text and visual prompt. For the text prompt, we propose a content guided refinement strategy that performs preliminary visual content analysis to generate structured information before answering the question. Additionally, we employ a visual attention refinement strategy that highlights question-relevant visual tokens to increase the model’s attention to visual content that aligns with the question. Extensive experiments demonstrate that these challenges can be significantly mitigated with our proposed dataset and techniques.

Composed Image Retrieval (CIR) is a vision-language task utilizing queries comprising images and textual descriptions to achieve precise image retrieval. This task seeks to find images that are visually similar to a reference image while incorporating specific changes or features described textually (visual delta). CIR enables a more flexible and user-specific retrieval by bridging visual data with verbal instructions. This paper introduces a novel generative method that augments Composed Image Retrieval by Composed Image Generation (CIG) to provide pseudo-target images. CIG utilizes a textual inversion network to map reference images into semantic word space, which generates pseudo-target images in combination with textual descriptions. These images serve as additional visual information, significantly improving the accuracy and relevance of retrieved images when integrated into existing retrieval frameworks. Experiments conducted across multiple CIR datasets and several baseline methods demonstrate improvements in retrieval performance, which shows the potential of our approach as an effective add-on for existing composed image retrieval. Project Page: https://lan-lw.github.io/CIG/

Temporal Domain Generalization (TDG) aims to generalize across temporal distribution shifts, e.g., lexical change over time.Prior work often addresses this by predicting future model weights.However, full model prediction is prohibitively expensive for even reasonably sized models.Thus, recent methods only predict the classifier layer, limiting generalization by failing to adjust other model components.To address this, we propose Temporal Experts Averaging (TEA), a novel and scalable TDG framework that updates the entire model using weight averaging to maximize generalization potential while minimizing computational costs.Our theoretical analysis guides us to two steps that enhance generalization to future domains.First, we create expert models with functional diversity yet parameter similarity by fine-tuning a domain-agnostic base model on individual temporal domains while constraining weight changes.Second, we optimize the bias-variance tradeoff through adaptive averaging coefficients derived from modeling temporal weight trajectories in a principal component subspace.Expert's contributions are based on their projected proximity to future domains.Extensive experiments across 7 TDG benchmarks, 5 models, and 2 TDG settings shows TEA outperforms prior TDG methods by up to 69% while being up to 60x more efficient 1 .

Modern retrieval systems often struggle with upgrading to new and more powerful models due to the incompatibility of embeddings between the old and new models. This necessitates a costly process known as backfilling, which involves re-computing the embeddings for a large number of data samples. In vision, Backward-compatible Training (BT) has been proposed to ensure that the new model aligns with the old model's embeddings. This paper extends the concept of vision-only BT to the field of cross-modal retrieval, marking the first attempt to address Cross-modal BT (XBT). Our goal is to achieve backward-compatibility between Vision-Language Pretraining (VLP) models, such as CLIP, for the cross-modal retrieval task. To address XBT challenges, we propose an efficient solution: a projection module that maps the new model's embeddings to those of the old model. This module, pretrained solely with text data, significantly reduces the number of image-text pairs required for XBT learning, and, once it is pretrained, it avoids using the old model during training. Furthermore, we utilize parameter-efficient training strategies that improve efficiency and preserve the off-the-shelf new model's knowledge by avoiding any modifications. Experimental results on cross-modal retrieval datasets demonstrate the effectiveness of XBT and its potential to enable backfill-free upgrades when a new VLP model emerges.

In this work, we present DreamDance, a novel method for animating human images using only skeleton pose sequences as conditional inputs. Existing approaches struggle with generating coherent, high-quality content in an efficient and user-friendly manner. Concretely, baseline methods relying on only 2D pose guidance lack the cues of 3D information, leading to suboptimal results, while methods using 3D representation as guidance achieve higher quality but involve a cumbersome and time-intensive process. To address these limitations, DreamDance enriches 3D geometry cues from 2D poses by introducing an efficient diffusion model, enabling high-quality human image animation with various guidance. Our key insight is that human images naturally exhibit multiple levels of correlation, progressing from coarse skeleton poses to fine-grained geometry cues, and further from these geometry cues to explicit appearance details. Capturing such correlations could enrich the guidance signals, facilitating intra-frame coherency and inter-frame consistency. Specifically, we construct the TikTok-Dance5K dataset, comprising 5K high-quality dance videos with detailed frame annotations, including human pose, depth, and normal maps. Next, we introduce a Mutually Aligned Geometry Diffusion Model to generate fine-grained depth and normal maps for enriched guidance. Finally, a Cross-domain Controller incorporates multi-level guidance to animate human images effectively with a video diffusion model. Extensive experiments demonstrate that our method achieves state-of-the-art performance in animating human images.

Supervised approaches for learning spatio-temporal scene graphs (STSG) from video are greatly hindered due to their reliance on STSG-annotated videos, which are labor-intensive to construct at scale. Is it feasible to instead use readily available video captions as weak supervision? To address this question, we propose LASER, a neuro-symbolic framework to enable training STSG generators using only video captions. LASER employs large language models to first extract logical specifications with rich spatio-temporal semantic information from video captions. LASER then trains the underlying STSG generator to align the predicted STSG with the specification. The alignment algorithm overcomes the challenges of weak supervision by leveraging a differentiable symbolic reasoner and using a combination of contrastive, temporal, and semantics losses. The overall approach efficiently trains low-level perception models to extract a fine-grained STSG that conforms to the video caption. In doing so, it enables a novel methodology for learning STSGs without tedious annotations. We evaluate our method on three video datasets: OpenPVSG, 20BN, and MUGEN. Our approach demonstrates substantial improvements over fully-supervised baselines, achieving a unary predicate prediction accuracy of 27.78% (+12.65%) and a binary recall@5 of 0.42 (+0.22) on OpenPVSG. Additionally, LASER exceeds baselines by 7% on 20BN and 5.2% on MUGEN in terms of overall predicate prediction accuracy.

We present a novel frequency-based Self-Supervised Learning (SSL) approach that significantly enhances its efficacy for pre-training. Prior work in this direction masks out pre-defined frequencies in the input image and employs a reconstruction loss to pre-train the model. While achieving promising results, such an implementation has two fundamental limitations as identified in our paper. First, using pre-defined frequencies overlooks the variability of image frequency responses. Second, pre-trained with frequency-filtered images, the resulting model needs relatively more data to adapt to naturally looking images during fine-tuning. To address these drawbacks, we propose FOurier transform compression with seLf-Knowledge distillation (FOLK), integrating two dedicated ideas. First, inspired by image compression, we adaptively select the masked-out frequencies based on image frequency responses, creating more suitable SSL tasks for pre-training. Second, we employ a two-branch framework empowered by knowledge distillation, enabling the model to take both the filtered and original images as input, largely reducing the burden of downstream tasks. Our experimental results demonstrate the effectiveness of FOLK in achieving competitive performance to many state-of-the-art SSL methods across various downstream tasks, including image classification, few-shot learning, and semantic segmentation.

Multimodal large language models (MLLMs) offer a powerful mechanism for interpreting visual information. However, they often suffer from hallucinations, which impede the real-world usage of these models. Existing methods attempt to alleviate this issue by designing special decoding strategies that penalize the summary tokens. However, these methods lack analysis of the relationship between hallucination and summarization mechanism of LLMs. Interestingly, we find that penalizing summary tokens is not necessary: merely intervening the query-key parameters variance, without costing extra inference time, still alleviates hallucinations. Specifically, we explore the causes of hallucinations by analyzing localized self-attention patterns called ``anchor" tokens and define the attention localization degree of the model as token propagation probabilities. Our analysis reveals that over-propagation of anchor tokens occurs when the distribution of eigenvalues of the query and key matrices has a non-zero mean and a polarized variance, leading to excessive dependence on anchor tokens while neglecting vision information and describes the image content with hallucination. Based on the observation, we propose a versatile plug-and-play decoding strategy, Dynamic Token Propagation Mechanism (TAME), to alleviate excessive propagation by dynamically intervening the eigenspectrum variance of the attention weight, thereby alleviating hallucinations without relying on complex decoding strategies. Extensive experiments reveal a correlation between the eigenspectrum and hallucinations across various MLLMs, and show that TAME reduces the percentage of hallucinated objects.

Neural Theorem Provers (NTPs) present a promising framework for neuro-symbolic reasoning, combining end-to-end differentiability with the interpretability of symbolic logic programming. However, optimizing NTPs remains a significant challenge due to their complex objective landscape and gradient sparcity. On the other hand, Knowledge Graph Embedding (KGE) methods offer smooth optimization with well-defined learning objectives but often lack interpretability. In this work, we propose several strategies to integrate the strengths of NTPs and KGEs, and demonstrate substantial improvements in both accuracy and computational efficiency. Specifically, we show that by leveraging the strength of structural learning in KGEs, we can greatly improve NTPs’ poorly structured embedding space, while by substituting NTPs with efficient KGE operations, we can significantly reduce evaluation time by over 1000$\times$ on large-scale dataset such as WN18RR with a mild accuracy trade-off.

Recent embodied agents are primarily built based on reinforcement learning (RL) or large language models (LLMs). Among them, RL agents are efficient for deployment but only perform very few tasks. By contrast, giant LLM agents (often more than 1000B parameters) present strong generalization while demanding enormous computing resources. In this work, we combine their advantages while avoiding the drawbacks by conducting the proposed referee RL on our developed large auto-regressive model (LARM). Specifically, LARM is built upon a lightweight LLM (fewer than 5B parameters) and directly outputs the next action to execute rather than text. We mathematically reveal that classic RL feedbacks vanish in long-horizon embodied exploration and introduce a giant LLM based referee to handle this reward vanishment during training LARM. In this way, LARM learns to complete diverse open-world tasks without human intervention. Especially, LARM successfully harvests enchanted diamond equipment in Minecraft, which demands significantly longer decision-making chains than the highest achievements of prior best methods.

Recent advancements in video generation have enabled the creation of high-quality, visually compelling videos. However, generating videos that adhere to the laws of physics remains a critical challenge for applications requiring realism and accuracy. In this work, we propose **PhysHPO**, a novel framework for Hierarchical Cross-Modal Direct Preference Optimization, to tackle this challenge by enabling fine-grained preference alignment for physically plausible video generation. PhysHPO optimizes video alignment across four hierarchical granularities: a) ***Instance Level***, aligning the overall video content with the input prompt; b) ***State Level***, ensuring temporal consistency using boundary frames as anchors; c) ***Motion Level***, modeling motion trajectories for realistic dynamics; and d) ***Semantic Level***, maintaining logical consistency between narrative and visuals. Recognizing that real-world videos are the best reflections of physical phenomena, we further introduce an automated data selection pipeline to efficiently identify and utilize *"good data"* from existing large-scale text-video datasets, thereby eliminating the need for costly and time-intensive dataset construction. Extensive experiments on both physics-focused and general capability benchmarks demonstrate that PhysHPO significantly improves physical plausibility and overall video generation quality of advanced models. To the best of our knowledge, this is the first work to explore fine-grained preference alignment and data selection for video generation, paving the way for more realistic and human-preferred video generation paradigms.

Multi-modal large language models (MLLMs) are making rapid progress toward general-purpose embodied agents. However, existing MLLMs do not reliably capture fine-grained links between low-level visual features and high-level textual semantics, leading to weak grounding and inaccurate perception. To overcome this challenge, we propose ESCA, a framework that contextualizes embodied agents by grounding their perception in spatial-temporal scene graphs. At its core is SGCLIP, a novel, open-domain, promptable foundation model for generating scene graphs that is based on CLIP. SGCLIP is trained on 87K+ open-domain videos using a neurosymbolic pipeline that aligns automatically generated captions with scene graphs produced by the model itself, eliminating the need for human-labeled annotations. We demonstrate that SGCLIP excels in both prompt-based inference and task-specific fine-tuning, achieving state-of-the-art results on scene graph generation and action localization benchmarks. ESCA with SGCLIP improves perception for embodied agents based on both open-source and commercial MLLMs, achieving state of-the-art performance across two embodied environments. Notably, ESCA significantly reduces agent perception errors and enables open-source models to surpass proprietary baselines. We release the source code for SGCLIP model training at https://github.com/video-fm/LASER and for the embodied agent at https://github.com/video-fm/ESCA.

Large Multimodal Models (LMMs) have achieved impressive progress in visual perception and reasoning. However, when confronted with visually ambiguous or non-semantic scene text, they often struggle to accurately spot and understand the content, frequently generating semantically plausible yet visually incorrect answers, which we refer to as semantic hallucination. In this work, we investigate the underlying causes of semantic hallucination and identify a key finding: Transformer layers in LLM with stronger attention focus on scene text regions are less prone to producing semantic hallucinations. Thus, we propose a training-free semantic hallucination mitigation framework comprising two key components: (1) ZoomText, a coarse-to-fine strategy that identifies potential text regions without external detectors; and (2) Grounded Layer Correction, which adaptively leverages the internal representations from layers less prone to hallucination to guide decoding, correcting hallucinated outputs for non-semantic samples while preserving the semantics of meaningful ones. To enable rigorous evaluation, we introduce TextHalu-Bench, a benchmark of 1,740 samples spanning both semantic and non-semantic cases, with manually curated question–answer pairs designed to probe model hallucinations. Extensive experiments demonstrate that our method not only effectively mitigates semantic hallucination but also achieves strong performance on public benchmarks for scene text spotting and understanding.

Backfilling is the process of re-extracting all gallery embeddings from upgraded models in image retrieval systems. It inevitably spends a prohibitively large amount of computational cost and even entails the downtime of the service. Although backward-compatible learning sidesteps this challenge by tackling query-side representations, this leads to suboptimal solutions in principle because gallery embeddings cannot benefit from model upgrades. We address this dilemma by introducing an online backfilling algorithm, which enables us to achieve a progressive performance improvement during the backfilling process without sacrificing the full performance of the new model after the completion of backfilling. To this end, we first show that a simple distance rank merge is a reasonable option for online backfilling. Then, we incorporate a reverse transformation module for more effective and efficient merging, which is further enhanced by adopting metric-compatible contrastive learning. These two components help to make the distances of old and new models compatible, resulting in desirable merge results during backfilling with no extra computational over-head. Extensive experiments show the benefit of our frame-work on four standard benchmarks in various settings.

Realizing unified 3D object detection, including both indoor and outdoor scenes, holds great importance in applications like robot navigation. However, involving various scenarios of data to train models poses challenges due to their significantly distinct characteristics, e.g., diverse geometry properties and heterogeneous domain distributions. In this work, we propose to address the challenges from two perspectives, the algorithm perspective and data perspective. In terms of the algorithm perspective, we first build a monocular 3D object detector based on the bird's-eye-view (BEV) detection paradigm, where the explicit feature projection is beneficial to addressing the geometry learning ambiguity. In this detector, we split the classical BEV detection architecture into two stages and propose an uneven BEV grid design to handle the convergence instability caused by geometry difference between scenarios. Besides, we develop a sparse BEV feature projection strategy to reduce the computational cost and a unified domain alignment method to handle heterogeneous domains. From the data perspective, we propose to incorporate depth information to improve training robustness. Specifically, we build the first unified multi-modal 3D object detection benchmark MM-Omni3D and extend the aforementioned monocular detector to its multi-modal version, which is the first unified multi-modal 3D object detector. We name the designed monocular and multi-modal detectors as UniMODE and MM-UniMODE, respectively. The experimental results reveal several insightful findings highlighting the benefits of multi-modal data and confirm the effectiveness of all the proposed strategies.

In this paper, we formally address universal object detection, which aims to detect every category in every scene. The dependence on human annotations, the limited visual information, and the novel categories in open world severely restrict the universality of detectors. We propose UniDetector, a universal object detector that recognizes enormous categories in the open world. The critical points for UniDetector are: 1) it leverages images of multiple sources and heterogeneous label spaces in training through image-text alignment, which guarantees sufficient information for universal representations. 2) it involves heterogeneous supervision training, which alleviates the dependence on the limited fully-labeled images. 3) it generalizes to open world easily while keeping the balance between seen and unseen classes. 4) it further promotes generalizing to novel categories through our proposed decoupling training manner and probability calibration. These contributions allow UniDetector to detect over 7 k categories, the largest measurable size so far, with only about 500 classes participating in training. Our UniDetector behaves the strong zero-shot ability on large-vocabulary datasets - it surpasses supervised baselines by more than 5% without seeing any corresponding images. On 13 detection datasets with various scenes, UniDetector also achieves state-of-the-art performance with only a 3% amount of training data.

With the success of large language models (LLMs), integrating the vision model into LLMs to build vision-language foundation models has gained much more interest recently. However, existing LLM-based large multimodal models (e.g., Video-LLaMA, VideoChat) can only take in a limited number of frames for short video understanding. In this study, we mainly focus on designing an efficient and effective model for long-term video understanding. Instead of trying to process more frames simultaneously like most existing work, we propose to process videos in an online manner and store past video information in a memory bank. This allows our model to reference historical video content for long-term analysis without exceeding LLMs' context length constraints or GPU memory limits. Our memory bank can be seamlessly integrated into current multimodal LLMs in an off-the-shelf manner. We conduct extensive experiments on various video understanding tasks, such as long-video understanding, video question answering, and video captioning, and our model can achieve state-of-the-art performances across multiple datasets.

Recent advances in instruction tuning have led to the development of State-of-the-Art Large Multimodal Models (LMMs). Given the novelty of these models, the impact of visual adversarial attacks on LMMs has not been thoroughly examined. We conduct a comprehensive study of the robustness of various LMMs against different adversarial attacks, evaluated across tasks including image classification, image captioning, and Visual Question Answer (VQA). We find that in general LMMs are not robust to visual adversarial inputs. However, our findings suggest that context provided to the model via prompts—such as questions in a QA pair—helps to mitigate the effects of visual adversarial inputs. Notably, the LMMs evaluated demonstrated remarkable resilience to such attacks on the ScienceQA task with only an 8.10% drop in performance compared to their visual counterparts which dropped 99.73%. We also propose a new approach to real-world image classification which we term query decomposition. By incorporating existence queries into our input prompt we observe diminished attack effectiveness and improvements in image classification accuracy. This research highlights a previously under explored facet of LMM robustness and sets the stage for future work aimed at strengthening the resilience of multimodal systems in adversarial environments.

Few-shot object detection (FSOD) aims to detect objects with only a few training examples. Visual feature extraction and query-support similarity learning are the two critical components. Existing works are usually developed based on ImageNet pre-trained vision backbones and design sophis-ticated metric-learning networks for few-shot learning, but still have inferior accuracy. In this work, we study few-shot object detection using modern foundation models. First, vision-only contrastive pre-trained DINOv2 model is used for the vision backbone, which shows strong transferable performance without tuning the parameters. Second, Large Language Model (LLM) is employed for contextualized few-shot learning with the input of all classes and query image proposals. Language instructions are carefully designed to prompt the LLM to classify each proposal in context. The contextual information include proposal-proposal relations, proposal-class relations, and class-class relations, which can largely promote few-shot learning. We comprehensively evaluate the proposed model (FM-FSOD) in multiple FSOD benchmarks, achieving state-of-the-arts performance.

Composed Image Retrieval (CIR) is a task that retrieves images similar to a query, based on a provided textual modification. Current techniques rely on supervised learning for CIR models using labeled triplets of the. These specific triplets are not as commonly available as simple image-text pairs, limiting the widespread use of CIR and its scalability. On the other hand, zero-shot CIR can be relatively easily trained with image-caption pairs without considering the image-to-image relation, but this approach tends to yield lower accuracy. We propose a new semi-supervised CIR approach where we search for a reference and its related target images in auxiliary data and learn our large language model-based Visual Delta Generator (VDG) to generate text describing the visual difference (i.e., visual delta) between the two. VDG, equipped with fluent language knowledge and being model agnostic, can generate pseudo triplets to boost the performance of CIR models. Our approach significantly improves the existing supervised learning approaches and achieves state-of-the-art results on the CIR benchmarks.

Realizing unified monocular 3D object detection, including both indoor and outdoor scenes, holds great importance in applications like robot navigation. However, involving various scenarios of data to train models poses challenges due to their significantly different characteristics, e.g., di-verse geometry properties and heterogeneous domain distributions. To address these challenges, we build a detector based on the bird's-eye-view (BEV) detection paradigm, where the explicit feature projection is beneficial to ad-dressing the geometry learning ambiguity when employing multiple scenarios of data to train detectors. Then, we split the classical BEV detection architecture into two stages and propose an uneven BEV grid design to handle the convergence instability caused by the aforementioned challenges. Moreover, we develop a sparse BEV feature projection strategy to reduce computational cost and a unified do-main alignment method to handle heterogeneous domains. Combining these techniques, a unified detector UniMODE is derived, which surpasses the previous state-of-the-art on the challenging Omni3D dataset (a large-scale dataset including both indoor and outdoor scenes) by 4.9% AP3D, revealing the first successful generalization of a BEV detector to unified 3D object detection.

We study the visual semantic embedding problem for image-text matching. Most existing work utilizes a tailored cross-attention mechanism to perform local alignment across the two image and text modalities. This is computationally expensive, even though it is more powerful than the unimodal dual-encoder approach. This work introduces a dual-encoder image-text matching model, leveraging a scene graph to represent captions with nodes for objects and attributes interconnected by relational edges. Utilizing a graph attention network, our model efficiently en-codes object-attribute and object-object semantic relations, resulting in a robust and fast-performing system. Representing caption as a scene graph offers the ability to utilize the strong relational inductive bias of graph neural networks to learn object-attribute and object-object relations effectively. To train the model, we propose losses that align the image and caption both at the holistic level (image-caption) and the local level (image-object entity), which we show is key to the success of the model. Our model is termed Composition model for Object Relations and Attributes, CORA. Experimental results on two prominent image-text retrieval benchmarks, Flickr30K and MS-COCO, demonstrate that CORA outperforms existing state-of-the-art computationally expensive cross-attention methods regarding recall score while achieving fast computation speed of the dual encoder. Our code is available at https://github.com/vkhoi/cora_cvpr24

The ability of large language models (LLMs) to process visual inputs has given rise to general-purpose vision systems, unifying various vision-language (VL) tasks by instruction tuning. However, due to the enormous diversity in input-output formats in the vision domain, existing general-purpose models fail to successfully integrate segmentation and multi-image inputs with coarse-level tasks into a single framework. In this work, we introduce VistaLLM, a powerful visual system that addresses coarse- and fine-grained VL tasks over single and multiple input images using a unified framework. VistaLLM utilizes an instruction-guided image tokenizer that filters global embeddings using task descriptions to extract compressed and refined features from numerous images. Moreover, VistaLLM employs a gradient-aware adaptive sampling technique to represent binary segmentation masks as sequences, significantly improving over previously used uniform sampling. To bolster the desired capability of VistaLLM, we curate CoinIt, a comprehensive coarse-to-fine instruction tuning dataset with 6.8M samples. We also address the lack of multi-image grounding datasets by introducing a novel task, AttCoSeg (Attribute-level Co-Segmentation), which boosts the model's reasoning and grounding capability over multiple input images. Extensive experiments on a wide range of V- and VL tasks demonstrate the effectiveness of VistaLLM by achieving consistent state-of-the-art performance over strong baselines across many downstream tasks. Our project page can be found at https://shramanpramanick.github.io/VistaLLM/.

We present an approach to pose object recognition as next token prediction. The idea is to apply a language decoder that auto-regressively predicts the text tokens from image embeddings to form labels. To ground this prediction process in auto-regression, we customize a non-causal attention mask for the decoder, incorporating two key features: modeling tokens from different labels to be independent, and treating image tokens as a prefix. This masking mechanism inspires an efficient method - one-shot sampling - to simultaneously sample tokens of multiple labels in parallel and rank generated labels by their probabilities during inference. To further enhance the efficiency, we propose a simple strategy to construct a compact decoder by simply discarding the intermediate blocks of a pretrained language model. This approach yields a decoder that matches the full model's performance while being notably more efficient. The code is available at github.com/kaiyuyue/nxtp.

Composing simple actions into complex actions is crucial yet challenging. Existing methods largely rely on language annotations to discern composable latent semantics, which is costly and labor-intensive. In this study, we introduce a novel framework to generate compositional actions without language auxiliaries. Our approach consists of three components: Action Coupling, Conditional Action Generation, and Decoupling Refinement. Action Coupling integrates two sub-actions to generate pseudo-training examples. Then, a conditional generative model, CVAE is employed to facilitate the diverse generation. Decoupling Refinement leverages a self-supervised pre-trained model MAE to ensure semantic consistency between sub-actions and compositional actions. Due to the lack of existing datasets containing both sub-actions and compositional actions, we create two new datasets, named HumanAct-C and UESTC-C. Both qualitative and quantitative assessments are conducted to show our efficacy.1

In the evolving landscape of video editing methodologies, a majority of deep learning techniques are often reliant on extensive datasets of observed input and ground truth sequence pairs for optimal performance. Such reliance often falters when acquiring data becomes challenging, especially in tasks like video dehazing and relighting, where replicating identical motions and camera angles in both corrupted and ground truth sequences is complicated. Moreover, these conventional methodologies perform best when the test distribution closely mirrors the training distribution. Recognizing these challenges, this paper introduces a novel video decomposition prior `VDP' framework which derives inspiration from professional video editing practices. Our methodology does not mandate task-specific external data corpus collection, instead pivots to utilizing the motion and appearance of the input video. VDP framework decomposes a video sequence into a set of multiple RGB layers and associated opacity levels. These set of layers are then manipulated individually to obtain the desired results. We addresses tasks such as video object segmentation, dehazing, and relighting. Moreover, we introduce a novel logarithmic video decomposition formulation for video relighting tasks, setting a new benchmark over the existing methodologies. We evaluate our approach on standard video datasets like DAVIS, REVIDE, & SDSD and show qualitative results on a diverse array of internet videos.

Online continual learning, the process of training models on streaming data, has gained increasing attention in recent years. However, a critical aspect often overlooked is the label delay, where new data may not be labeled due to slow and costly annotation processes. We introduce a new continual learning framework with explicit modeling of the label delay between data and label streams over time steps. In each step, the framework reveals both unlabeled data from the current time step t and labels delayed with d steps, from the time step t−d. In our extensive experiments amounting to 1060 GPU days, we show that merely augmenting the computational resources is insufficient to tackle this challenge. Our findings underline a notable performance decline when solely relying on labeled data when the label delay becomes significant. More surprisingly, when using state-of-the-art SSL and TTA techniques to utilize the newer, unlabeled data, they fail to surpass the performance of a naïve method that simply trains on the delayed supervised stream. To this end, we introduce a simple, efficient baseline that rehearses from the labeled memory samples that are most similar to the new unlabeled samples. This method bridges the accuracy gap caused by label delay without significantly increasing computational complexity. We show experimentally that our method is the least affected by the label delay factor and in some cases successfully recovers the accuracy of the non-delayed counterpart. We conduct various ablations and sensitivity experiments, demonstrating the effectiveness of our approach.

Illustration is a fundamental mode of human expression and communication. Certain types of motion that accompany speech can provide this illustrative mode of communication. While Augmented and Virtual Reality technologies (AR/VR) have introduced tools for producing drawings with hand motions (air drawing), they typically require costly hardware and additional digital markers, thereby limiting their accessibility and portability. Furthermore, air drawing demands considerable skill to achieve aesthetic results. To address these challenges, we introduce the concept of AirSketch, aimed at generating faithful and visually coherent sketches directly from hand motions, eliminating the need for complicated headsets or markers. We devise a simple augmentation-based self-supervised training procedure, enabling a controllable image diffusion model to learn to translate from highly noisy hand tracking images to clean, aesthetically pleasing sketches, while preserving the essential visual cues from the original tracking data. We present two air drawing datasets to study this problem. Our findings demonstrate that beyond producing photo-realistic images from precise spatial inputs, controllable image diffusion can effectively produce a refined, clear sketch from a noisy input. Our work serves as an initial step towards marker-less air drawing and reveals distinct applications of controllable diffusion models to AirSketch and AR/VR in general.

It is now well known that neural networks can be wrong with high confidence in their predictions, leading to poor calibration. The most common post-hoc approach to compensate for this is to perform temperature scaling, which adjusts the confidences of the predictions on any input by scaling the logits by a fixed value. Whilst this approach typically improves the average calibration across the whole test dataset, this improvement typically reduces the individual confidences of the predictions irrespective of whether the classification of a given input is correct or incorrect. With this insight, we base our method on the observation that different samples contribute to the calibration error by varying amounts, with some needing to increase their confidence and others needing to decrease it. Therefore, for each input, we propose to predict a different temperature value, allowing us to adjust the mismatch between confidence and accuracy at a finer granularity. Our method is applied post-hoc, enabling it to be very fast with a negligible memory footprint and is applied to off-the-shelf pre-trained classifiers. We test our method on the ResNet50 and WideResNet28-10 architectures using the CIFAR10/100 and Tiny-ImageNet datasets, showing that producing per-data-point temperatures improves the expected calibration error across the whole test set.

Implicit neural representations (INR) have gained increasing attention in representing 3D scenes and images, and have been recently applied to encode videos (e.g., NeRV [1], E-NeRV [2]). While achieving promising results, existing INR-based methods are limited to encoding a handful of short videos (e.g., seven 5-second videos in the UVG dataset) with redundant visual content, leading to a model design that fits individual video frames independently and is not efficiently scalable to a large number of diverse videos. This paper focuses on developing neural representations for a more practical setup - encoding long and/or a large number of videos with diverse visual content. We first show that instead of dividing videos into small subsets and encoding them with separate models, encoding long and diverse videos jointly with a unified model achieves better compression results. Based on this observation, we propose D-NeRV, a novel neural representation framework designed to encode diverse videos by (i) decoupling clip-specific visual content from motion information, (ii) introducing temporal reasoning into the implicit neural network, and (iii) employing the task-oriented flow as intermediate output to reduce spatial redundancies. Our new model largely surpasses NeRV and traditional video compression techniques on UCF101 and UVG datasets on the video compression task. Moreover, when used as an efficient data-loader, D-NeRV achieves 3%-10% higher accuracy than NeRV on action recognition tasks on the UCF101 dataset under the same compression ratios.

In this paper, we formally address universal object detection, which aims to detect every scene and predict every category. The dependence on human annotations, the limited visual information, and the novel categories in the open world severely restrict the universality of traditional detectors. We propose UniDetector, a universal object detector that has the ability to recognize enormous categories in the open world. The critical points for the universality of UniDetector are: 1) it leverages images of multiple sources and heterogeneous label spaces for training through the alignment of image and text spaces, which guarantees sufficient information for universal representations. 2) it generalizes to the open world easily while keeping the balance between seen and unseen classes, thanks to abundant information from both vision and language modalities. 3) it further promotes the generalization ability to novel categories through our proposed decoupling training manner and probability calibration. These contributions allow UniDetector to detect over 7k categories, the largest measurable category size so far, with only about 500 classes participating in training. Our UniDetector behaves the strong zero-shot generalization ability on largevocabulary datasets - it surpasses the traditional supervised baselines by more than 4% on average without seeing any corresponding images. On 13 public detection datasets with various scenes, UniDetector also achieves state-of-the-art performance with only a 3% amount of training data.11Codes are available at https://github.com/zhenyuw16/UniDetector.

Implicit neural representations store videos as neural networks and have performed well for various vision tasks such as video compression and denoising. With frame index or positional index as input, implicit representations (NeRV, E-NeRV, etc.) reconstruct video frames from fixed and content-agnostic embeddings. Such embedding largely limits the regression capacity and internal generalization for video interpolation. In this paper, we propose a Hybrid Neural Representation for Videos (HNeRV), where a learnable encoder generates content-adaptive embeddings, which act as the decoder input. Besides the input embedding, we introduce HNeRV blocks, which ensure model parameters are evenly distributed across the entire network, such that higher layers (layers near the output) can have more capacity to store high-resolution content and video details. With content-adaptive embeddings and redesigned architecture, HNeRV outperforms implicit methods in video regression tasks for both reconstruction quality (+4.7 PSNR) and convergence speed (16 × faster), and shows better internal generalization. As a simple and efficient video representation, HNeRV also shows decoding advantages for speed, flexibility, and deployment, compared to traditional codecs (H.264, H.265) and learning-based compression methods. Finally, we explore the effectiveness of HNeRV on downstream tasks such as video compression and video inpainting.

We introduce Patch Aligned Contrastive Learning (PACL), a modified compatibility function for CLIP's contrastive loss, intending to train an alignment between the patch tokens of the vision encoder and the CLS token of the text encoder. With such an alignment, a model can identify regions of an image corresponding to a given text input, and therefore transfer seamlessly to the task of open vocabulary semantic segmentation without requiring any segmentation annotations during training. Using pre-trained CLIP encoders with PACL, we are able to set the state-of-the-art on the task of open vocabulary zero-shot segmentation on 4 different segmentation benchmarks: Pascal VOC, Pascal Context, COCO Stuff and ADE20K. Furthermore, we show that PACL is also applicable to image-level predictions and when used with a CLIP backbone, provides a general improvement in zero-shot classification accuracy compared to CLIP, across a suite of 12 image classification datasets.

When deploying a machine learning model to a new environment, we often encounter the distribution shift problem - meaning the target data distribution is different from the model's training distribution. In this paper, we assume that labels are not provided for this new domain, and that we do not store the source data (e.g., for privacy reasons). It has been shown that even small shifts in the data distribution can affect the model's performance severely. Test Time Adaptation offers a means to combat this problem, as it allows the model to adapt during test time to the new data distribution, using only unlabeled test data batches. To achieve this, the predominant approach is to optimize a surrogate loss on the test-time unlabeled target data. In particular, minimizing the prediction's entropy on target samples [34] has received much interest as it is task-agnostic and does not require altering the model's training phase (e.g., does not require adding a self-supervised task during training on the source domain). However, as the target data's batch size is often small in real-world scenarios (e.g., autonomous driving models process each few frames in real-time), we argue that this surrogate loss is not optimal since it often collapses with small batch sizes. To tackle this problem, in this paper, we propose to use an invariance regularizer as the surrogate loss during test-time adaptation, motivated by our theoretical results regarding the model's performance under input transformations. The resulting method (TIPI-Test time adaPtation with transformation Invariance) is validated with extensive experiments in various benchmarks (Cifar10-C, Cifar100-C, ImageNet-C, DIGITS, and VisDA17). Remarkably, TIPI is robust against small batch sizes (as small as 2 in our experiments), and consistently outperforms TENT [34] in all settings. Our code is released at https://github.com/atuannguyen/TIPI.

Continual Learning (CL) aims to sequentially train models on streams of incoming data that vary in distribution by preserving previous knowledge while adapting to new data. Current CL literature focuses on restricted access to previously seen data, while imposing no constraints on the computational budget for training. This is unreasonable for applications in-the-wild, where systems are primarily constrained by computational and time budgets, not storage. We revisit this problem with a large-scale benchmark and analyze the performance of traditional CL approaches in a compute-constrained setting, where effective memory samples used in training can be implicitly restricted as a consequence of limited computation. We conduct experiments evaluating various CL sampling strategies, distillation losses, and partial fine-tuning on two large-scale datasets, namely ImageNet2K and Continual Google Landmarks V2 in data incremental, class incremental, and time incremental settings. Through extensive experiments amounting to a total of over 1500 GPU-hours, we find that, under compute-constrained setting, traditional CL approaches, with no exception, fail to outperform a simple minimal baseline that samples uniformly from memory. Our conclusions are consistent in a different number of stream time steps, e.g., 20 to 200, and under several computational budgets. This suggests that most existing CL methods are particularly too computationally expensive for realistic budgeted deployment. Code for this project is available at: https://github.com/drimpossible/BudgetCL.

Open-vocabulary image segmentation is attracting increasing attention due to its critical applications in the real world. Traditional closed-vocabulary segmentation methods are not able to characterize novel objects, whereas several recent open-vocabulary attempts obtain unsatisfactory results, i.e., notable performance reduction on the closedvocabulary and massive demand for extra data. To this end, we propose OPSNet, an omnipotent and data-efficient framework for Open-vocabulary Panoptic Segmentation. Specifically, the exquisitely designed Embedding Modulation module, together with several meticulous components, enables adequate embedding enhancement and information exchange between the segmentation model and the visual-linguistic well-aligned CLIP encoder, resulting in superior segmentation performance under both open- and closed-vocabulary settings with much fewer need of additional data. Extensive experimental evaluations are conducted across multiple datasets (e.g., COCO, ADE20K, Cityscapes, and PascalContext) under various circumstances, where the proposed OPSNet achieves state-of-theart results, which demonstrates the effectiveness and generality of the proposed approach. The project page is https://opsnet-page.github.io.

We revisit the common practice of evaluating adaptation of Online Continual Learning (OCL) algorithms through the metric of online accuracy, which measures the accuracy of the model on the immediate next few samples. However, we show that this metric is unreliable, as even vacuous blind classifiers, which do not use input images for prediction, can achieve unrealistically high online accuracy by exploiting spurious label correlations in the data stream. Our study reveals that existing OCL algorithms can also achieve high online accuracy, but perform poorly in retaining useful information, suggesting that they unintentionally learn spurious label correlations. To address this issue, we propose a novel metric for measuring adaptation based on the accuracy on the near-future samples, where spurious correlations are removed. We benchmark existing OCL approaches using our proposed metric on large-scale datasets under various computational budgets and find that better generalization can be achieved by retaining and reusing past seen information. We believe that our proposed metric can aid in the development of truly adaptive OCL methods. We provide code to reproduce our results at https://github.com/drimpossible/EvalOCL.

Modern retrieval system often requires recomputing the representation of every piece of data in the gallery when updating to a better representation model. This process is known as backfilling and can be especially costly in the real world where the gallery often contains billions of samples. Recently, researchers have proposed the idea of Backward Compatible Training (BCT) where the new representation model can be trained with an auxiliary loss to make it backward compatible with the old representation. In this way, the new representation can be directly compared with the old representation, in principle avoiding the need for any backfilling. However, follow-up work shows that there is an inherent trade-off where a backward compatible representation model cannot simultaneously maintain the performance of the new model itself. This paper reports our "not-so-surprising" finding that adding extra dimensions to the representation can help here. However, we also found that naively increasing the dimension of the representation did not work. To deal with this, we propose Backward-compatible Training with a novel Basis Transformation (BT2). A basis transformation (BT) is basically a learnable set of parameters that applies an orthonormal transformation. Such a transformation possesses an important property whereby the original information contained in its input is retained in its output. We show in this paper how a BT can be utilized to add only the necessary amount of additional dimensions. We empirically verify the advantage of BT2over other state-of-the-art methods in a wide range of settings. We then further extend BT2to other challenging yet more practical settings, including significant changes in model architecture (CNN to Transformers), modality change, and even a series of updates in the model architecture mimicking the evolution of deep learning models in the past decade. Our code is available at https://github.com/YifeiZhou02/BT-2.

Transformers for graph data are increasingly widely studied and successful in numerous learning tasks. Graph inductive biases are crucial for Graph Transformers, and previous works incorporate them using message-passing modules and/or positional encodings. However, Graph Transformers that use message-passing inherit known issues of message-passing, and differ significantly from Transformers used in other domains, thus making transfer of research advances more difficult. On the other hand, Graph Transformers without message-passing often perform poorly on smaller datasets, where inductive biases are more crucial. To bridge this gap, we propose the Graph Inductive bias Transformer (GRIT) --- a new Graph Transformer that incorporates graph inductive biases without using message passing. GRIT is based on several architectural changes that are each theoretically and empirically justified, including: learned relative positional encodings initialized with random walk probabilities, a flexible attention mechanism that updates node and node-pair representations, and injection of degree information in each layer. We prove that GRIT is expressive --- it can express shortest path distances and various graph propagation matrices. GRIT achieves state-of-the-art empirical performance across a variety of graph datasets, thus showing the power that Graph Transformers without message-passing can deliver.

Recent methods in geometric deep learning have introduced various neural networks to operate over data that lie on Riemannian manifolds. Such networks are often necessary to learn well over graphs with a hierarchical structure or to learn over manifold-valued data encountered in the natural sciences. These networks are often inspired by and directly generalize standard Euclidean neural networks. However, extending Euclidean networks is difficult and has only been done for a select few manifolds. In this work, we examine the residual neural network (ResNet) and show how to extend this construction to general Riemannian manifolds in a geometrically principled manner. Originally introduced to help solve the vanishing gradient problem, ResNets have become ubiquitous in machine learning due to their beneficial learning properties, excellent empirical results, and easy-to-incorporate nature when building varied neural networks. We find that our Riemannian ResNets mirror these desirable properties: when compared to existing manifold neural networks designed to learn over hyperbolic space and the manifold of symmetric positive definite matrices, we outperform both kinds of networks in terms of relevant testing metrics and training dynamics.

In this paper, we present a novel robust framework for low-level vision tasks, including denoising, object removal, frame interpolation, and super-resolution, that does not require any external training data corpus. Our proposed approach directly learns the weights of neural modules by optimizing over the corrupted test sequence, leveraging the spatio-temporal coherence and internal statistics of videos. Furthermore, we introduce a novel spatial pyramid loss that leverages the property of spatio-temporal patch recurrence in a video across the different scales of the video. This loss enhances robustness to unstructured noise in both the spatial and temporal domains. This further results in our framework being highly robust to degradation in input frames and yields state-of-the-art results on downstream tasks such as denoising, object removal, and frame interpolation. To validate the effectiveness of our approach, we conduct qualitative and quantitative evaluations on standard video datasets such as DAVIS, UCF-101, and VIMEO90K-T.

Advances in the field of visual-language contrastive learning have made it possible for many downstream applications to be carried out efficiently and accurately by simply taking the dot product between image and text representations. One of the most representative approaches proposed recently known as CLIP has quickly garnered widespread adoption due to its effectiveness. CLIP is trained with an InfoNCE loss that takes into account both positive and negative samples to help learn a much more robust representation space. This paper however reveals that the common downstream practice of taking a dot product is only a zeroth-order approximation of the optimization goal, resulting in a loss of information during test-time. Intuitively, since the model has been optimized based on the InfoNCE loss, test-time procedures should ideally also be in alignment. The question lies in how one can retrieve any semblance of negative samples information during inference in a computationally efficient way. We propose Distribution Normalization (DN), where we approximate the mean representation of a batch of test samples and use such a mean to represent what would be analogous to negative samples in the InfoNCE loss. DN requires no retraining or fine-tuning and can be effortlessly applied during inference. Extensive experiments on a wide variety of downstream tasks exhibit a clear advantage of DN over the dot product on top of other existing test-time augmentation methods.

Built on top of self-attention mechanisms, vision transformers have demonstrated remarkable performance on a variety of tasks recently. While achieving excellent performance, they still require relatively intensive computational cost that scales up drastically as the numbers of patches, self-attention heads and transformer blocks increase. In this paper, we argue that due to the large variations among images, their need for modeling long-range dependencies between patches differ. To this end, we introduce AdaViT, an adaptive computation framework that learns to derive usage policies on which patches, self-attention heads and transformer blocks to use throughout the backbone on a per-input basis, aiming to improve inference efficiency of vision transformers with a minimal drop of accuracy for image recognition. Optimized jointly with a transformer backbone in an end-to-end manner, a light-weight decision network is attached to the backbone to produce decisions on-the-fly. Extensive experiments on ImageNet demonstrate that our method obtains more than 2 × improvement on efficiency compared to state-of-the-art vision transformers with only 0.8% drop of accuracy, achieving good efficiency/accuracy trade-offs conditioned on different computational budgets. We further conduct quantitative and qualitative analysis on learned usage polices and provide more insights on the redundancy in vision transformers. Code is available at ht tps: / / gi thub. com/MengLcool/AdaVi T.

Recent advances in image editing techniques have posed serious challenges to the trustworthiness of multimedia data, which drives the research of image tampering detection. In this paper, we propose ObjectFormer to detect and localize image manipulations. To capture subtle manipulation traces that are no longer visible in the RGB domain, we extract high-frequency features of the images and combine them with RGB features as multimodal patch embeddings. Additionally, we use a set of learnable object prototypes as mid-level representations to model the object-level consistencies among different regions, which are further used to refine patch embeddings to capture the patch-level consistencies. We conduct extensive experiments on various datasets and the results verify the effectiveness of the proposed method, outperforming state-of-the-art tampering detection and localization methods.

The widespread dissemination of Deepfakes demands effective approaches that can detect perceptually convincing forged images. In this paper, we aim to capture the subtle manipulation artifacts at different scales using transformer models. In particular, we introduce a Multi-modal Multi-scale TRansformer (M2TR), which operates on patches of different sizes to detect local inconsistencies in images at different spatial levels. M2TR further learns to detect forgery artifacts in the frequency domain to complement RGB information through a carefully designed cross modality fusion block. In addition, to stimulate Deepfake detection research, we introduce a high-quality Deepfake dataset, SR-DF, which consists of 4,000 DeepFake videos generated by state-of-the-art face swapping and facial reenactment methods. We conduct extensive experiments to verify the effectiveness of the proposed method, which outperforms state-of-the-art Deepfake detection methods by clear margins.

Federated Learning (FL) refers to the decentralized and privacy-preserving machine learning framework in which multiple clients collaborate (with the help of a central server) to train a global model without sharing their data. However, most existing FL methods only focus on maximizing the model's performance on the source clients' data (e.g., mobile users) without considering its generalization ability to unknown target data (e.g., a new user). In this paper, we incorporate the problem of Domain Generalization (DG) into Federated Learning to tackle the aforementioned issue. However, virtually all existing DG methods require a centralized setting where data is shared across the domains, which violates the principles of decentralized FL and hence not applicable. To this end, we propose a simple yet novel representation learning framework, namely FedSR, which enables domain generalization while still respecting the decentralized and privacy-preserving natures of this FL setting. Motivated by classical machine learning algorithms, we aim to learn a simple representation of the data for better generalization. In particular, we enforce an L2-norm regularizer on the representation and a conditional mutual information (between the representation and the data given the label) regularizer to encourage the model to only learn essential information (while ignoring spurious correlations such as the background). Furthermore, we provide theoretical connections between the above two objectives and representation alignment in domain generalization. Extensive experimental results suggest that our method significantly outperforms relevant baselines in this particular problem.

We show that the effectiveness of the well celebrated Mixup can be further improved if instead of using it as the sole learning objective, it is utilized as an additional regularizer to the standard cross-entropy loss. This simple change not only improves accuracy but also significantly improves the quality of the predictive uncertainty estimation of Mixup in most cases under various forms of covariate shifts and out-of-distribution detection experiments. In fact, we observe that Mixup otherwise yields much degraded performance on detecting out-of-distribution samples possibly, as we show empirically, due to its tendency to learn models exhibiting high-entropy throughout; making it difficult to differentiate in-distribution samples from out-of-distribution ones. To show the efficacy of our approach (RegMixup), we provide thorough analyses and experiments on vision datasets (ImageNet & CIFAR-10/100) and compare it with a suite of recent approaches for reliable uncertainty estimation.

Recent progress in vision Transformers exhibits great success in various tasks driven by the new spatial modeling mechanism based on dot-product self-attention. In this paper, we show that the key ingredients behind the vision Transformers, namely input-adaptive, long-range and high-order spatial interactions, can also be efficiently implemented with a convolution-based framework. We present the Recursive Gated Convolution ($\textit{g}^\textit{n}$Conv) that performs high-order spatial interactions with gated convolutions and recursive designs. The new operation is highly flexible and customizable, which is compatible with various variants of convolution and extends the two-order interactions in self-attention to arbitrary orders without introducing significant extra computation. $\textit{g}^\textit{n}$Conv can serve as a plug-and-play module to improve various vision Transformers and convolution-based models. Based on the operation, we construct a new family of generic vision backbones named HorNet. Extensive experiments on ImageNet classification, COCO object detection and ADE20K semantic segmentation show HorNet outperform Swin Transformers and ConvNeXt by a significant margin with similar overall architecture and training configurations. HorNet also shows favorable scalability to more training data and larger model sizes. Apart from the effectiveness in visual encoders, we also show $\textit{g}^\textit{n}$Conv can be applied to task-specific decoders and consistently improve dense prediction performance with less computation. Our results demonstrate that $\textit{g}^\textit{n}$Conv can be a new basic module for visual modeling that effectively combines the merits of both vision Transformers and CNNs. Code is available at https://github.com/raoyongming/HorNet.

We present Spartan, a method for training sparse neural network models with a predetermined level of sparsity. Spartan is based on a combination of two techniques: (1) soft top-k masking of low-magnitude parameters via a regularized optimal transportation problem and (2) dual averaging-based parameter updates with hard sparsification in the forward pass. This scheme realizes an exploration-exploitation tradeoff: early in training, the learner is able to explore various sparsity patterns, and as the soft top-k approximation is gradually sharpened over the course of training, the balance shifts towards parameter optimization with respect to a fixed sparsity mask. Spartan is sufficiently flexible to accommodate a variety of sparsity allocation policies, including both unstructured and block-structured sparsity, global and per-layer sparsity budgets, as well as general cost-sensitive sparsity allocation mediated by linear models of per-parameter costs. On ImageNet-1K classification, we demonstrate that training with Spartan yields 95% sparse ResNet-50 models and 90% block sparse ViT-B/16 models while incurring absolute top-1 accuracy losses of less than 1% compared to fully dense training.

The ability to detect anomaly has long been recognized as an inherent human ability, yet to date, practical AI solutions to mimic such capability have been lacking. This lack of progress can be attributed to several factors. To begin with, the distribution of ``abnormalities'' is intractable. Anything outside of a given normal population is by definition an anomaly. This explains why a large volume of work in this area has been dedicated to modeling the normal distribution of a given task followed by detecting deviations from it. This direction is however unsatisfying as it would require modeling the normal distribution of every task that comes along, which includes tedious data collection. In this paper, we report our work aiming to handle these issues. To deal with the intractability of abnormal distribution, we leverage Energy Based Model (EBM). EBMs learn to associates low energies to correct values and higher energies to incorrect values. At its core, the EBM employs Langevin Dynamics (LD) in generating these incorrect samples based on an iterative optimization procedure, alleviating the intractable problem of modeling the world of anomalies. Then, in order to avoid training an anomaly detector for every task, we utilize an adaptive sparse coding layer. Our intention is to design a plug and play feature that can be used to quickly update what is normal during inference time. Lastly, to avoid tedious data collection, this mentioned update of the sparse coding layer needs to be achievable with just a few shots. Here, we employ a meta learning scheme that simulates such a few shot setting during training. We support our findings with strong empirical evidence.

Paraphrase Identification is a fundamental task in Natural Language Processing. While much progress has been made in the field, the performance of many state-of- the-art models often suffer from distribution shift during inference time. We verify that a major source of this performance drop comes from biases introduced by negative examples. To overcome these biases, we propose in this paper to train two separate models, one that only utilizes the positive pairs and the other the negative pairs. This enables us the option of deciding how much to utilize the negative model, for which we introduce a perplexity based out-of-distribution metric that we show can effectively and automatically determine how much weight it should be given during inference. We support our findings with strong empirical results.

Detecting spliced images is one of the emerging challenges in computer vision. Unlike prior methods that focus on detecting low-level artifacts generated during the manipulation process, we use an image retrieval approach to tackle this problem. When given a spliced query image, our goal is to retrieve the original image from a database of authentic images. To achieve this goal, we propose representing an image by its constituent objects based on the intuition that the finest granularity of manipulations is oftentimes at the object-level. We introduce a framework, object embeddings for spliced image retrieval (OE-SIR), that utilizes modern object detectors to localize object regions. Each region is then embedded and collectively used to represent the image. Further, we propose a student-teacher training paradigm for learning discriminative embeddings within object regions to avoid expensive multiple forward passes. Detailed analysis of the efficacy of different feature embedding models is also provided in this study. Extensive experimental results show that the OE-SIR achieves state-of-the-art performance in spliced image retrieval.

An image is worth a thousand words, conveying information that goes beyond the mere visual content therein. In this paper, we study the intent behind social media images with an aim to analyze how visual information can facilitate recognition of human intent. Towards this goal, we introduce an intent dataset, Intentonomy, comprising 14K images covering a wide range of everyday scenes. These images are manually annotated with 28 intent categories derived from a social psychology taxonomy. We then systematically study whether, and to what extent, commonly used visual information, i.e., object and context, contribute to human motive understanding. Based on our findings, we conduct further study to quantify the effect of attending to object and context classes as well as textual information in the form of hashtags when training an intent classifier. Our results quantitatively and qualitatively shed light on how visual and textual information can produce observable effects when predicting intent.1

The moments (a.k.a., mean and standard deviation) of latent features are often removed as noise when training image recognition models, to increase stability and reduce training time. However, in the field of image generation, the moments play a much more central role. Studies have shown that the moments extracted from instance normalization and positional normalization can roughly capture style and shape information of an image. Instead of being discarded, these moments are instrumental to the generation process. In this paper we propose Moment Exchange, an implicit data augmentation method that encourages the model to utilize the moment information also for recognition models. Specifically, we replace the moments of the learned features of one training image by those of another, and also interpolate the target labels—forcing the model to extract training signal from the moments in addition to the normalized features. As our approach is fast, operates entirely in feature space, and mixes different signals than prior methods, one can effectively combine it with existing augmentation approaches. We demonstrate its efficacy across several recognition benchmark data sets where it improves the generalization capability of highly competitive baseline networks with remarkable consistency.

Visual engagement in social media platforms comprises interactions with photo posts including comments, shares, and likes. In this paper, we leverage such Visual Engagement clues as supervisory signals for representation learning. However, learning from engagement signals is non-trivial as it is not clear how to bridge the gap between low-level visual information and high-level social interactions. We present VisE, a weakly supervised learning approach, which maps social images to pseudo labels derived by clustered engagement signals. We then study how models trained in this way benefit subjective downstream computer vision tasks such as emotion recognition or political bias detection. Through extensive studies, we empirically demonstrate the effectiveness of VisE across a diverse set of classification tasks beyond the scope of conventional recognition1.

While deep neural networks have achieved remarkable success in various computer vision tasks, they often fail to generalize to new domains and subtle variations of input images. Several defenses have been proposed to improve the robustness against these variations. However, current defenses can only withstand the specific attack used in training, and the models often remain vulnerable to other input variations. Moreover, these methods often degrade performance of the model on clean images and do not generalize to out-of-domain samples. In this paper we present Generative Adversarial Training, an approach to simultaneously improve the model’s generalization to the test set and out-of-domain samples as well as its robustness to unseen adversarial attacks. Instead of altering a low-level pre-defined aspect of images, we generate a spectrum of low-level, mid-level and high-level changes using generative models with a disentangled latent space. Adversarial training with these examples enable the model to withstand a wide range of attacks by observing a variety of input alterations during training. We show that our approach not only improves performance of the model on clean images and out-of-domain samples but also makes it robust against unforeseen attacks and outperforms prior work. We validate effectiveness of our method by demonstrating results on various tasks such as classification, segmentation and object detection.

Semi-supervised domain adaptation (SSDA) aims to adapt models trained from a labeled source domain to a different but related target domain, from which unlabeled data and a small set of labeled data are provided. Current methods that treat source and target supervision without distinction overlook their inherent discrepancy, resulting in a source-dominated model that has not effectively use the target supervision. In this paper, we argue that the labeled target data needs to be distinguished for effective SSDA, and propose to explicitly decompose the SSDA task into two sub-tasks: a semi-supervised learning (SSL) task in the target domain and an unsupervised domain adaptation (UDA) task across domains. By doing so, the two sub-tasks can better leverage the corresponding supervision and thus yield very different classifiers. To integrate the strengths of the two classifiers, we apply the well established co-training framework, in which the two classifiers exchange their high confident predictions to iteratively "teach each other" so that both classifiers can excel in the target domain. We call our approach Deep Co-training with Task decomposition (DeCoTa). DeCoTa requires no adversarial training and is easy to implement. Moreover, DeCoTa is well founded on the theoretical condition of when co-training would succeed. As a result, DeCoTa achieves state-of-the-art results on several SSDA datasets, outperforming the prior art by a notable 4% margin on DomainNet. Code is available at https://github.com/LoyoYang/DeCoTa.

Deepfakes ("deep learning" + "fake") are videos synthetically generated with AI algorithms. While they could be entertaining, they could also be misused for falsifying speeches and spreading misinformation. The process to create deepfakes involves both visual and auditory manipulations. Exploration on detecting visual deepfakes has produced a number of detection methods as well as datasets, while audio deepfakes (e.g. synthetic speech from text-to-speech or voice conversion systems) and the relationship between the video and audio modalities have been relatively neglected. In this work, we propose a novel visual / auditory deepfake joint detection task and show that exploiting the intrinsic synchronization between the visual and auditory modalities could benefit deepfake detection. Experiments demonstrate that the proposed joint detection framework outperforms independently trained models, and at the same time, yields superior generalization capability on unseen types of deepfakes.

We propose a novel neural representation for videos (NeRV) which encodes videos in neural networks. Unlike conventional representations that treat videos as frame sequences, we represent videos as neural networks taking frame index as input. Given a frame index, NeRV outputs the corresponding RGB image. Video encoding in NeRV is simply fitting a neural network to video frames and decoding process is a simple feedforward operation. As an image-wise implicit representation, NeRV output the whole image and shows great efficiency compared to pixel-wise implicit representation, improving the encoding speed by $\textbf{25}\times$ to $\textbf{70}\times$, the decoding speed by $\textbf{38}\times$ to $\textbf{132}\times$, while achieving better video quality. With such a representation, we can treat videos as neural networks, simplifying several video-related tasks. For example, conventional video compression methods are restricted by a long and complex pipeline, specifically designed for the task. In contrast, with NeRV, we can use any neural network compression method as a proxy for video compression, and achieve comparable performance to traditional frame-based video compression approaches (H.264, HEVC \etc). Besides compression, we demonstrate the generalization of NeRV for video denoising. The source code and pre-trained model can be found at https://github.com/haochen-rye/NeRV.git.

Tractably modelling distributions over manifolds has long been an important goal in the natural sciences. Recent work has focused on developing general machine learning models to learn such distributions. However, for many applications these distributions must respect manifold symmetries—a trait which most previous models disregard. In this paper, we lay the theoretical foundations for learning symmetry-invariant distributions on arbitrary manifolds via equivariant manifold flows. We demonstrate the utility of our approach by learning quantum field theory-motivated invariant SU(n) densities and by correcting meteor impact dataset bias.

Many widely used datasets for graph machine learning tasks have generally been homophilous, where nodes with similar labels connect to each other. Recently, new Graph Neural Networks (GNNs) have been developed that move beyond the homophily regime; however, their evaluation has often been conducted on small graphs with limited application domains. We collect and introduce diverse non-homophilous datasets from a variety of application areas that have up to 384x more nodes and 1398x more edges than prior datasets. We further show that existing scalable graph learning and graph minibatching techniques lead to performance degradation on these non-homophilous datasets, thus highlighting the need for further work on scalable non-homophilous methods. To address these concerns, we introduce LINKX --- a strong simple method that admits straightforward minibatch training and inference. Extensive experimental results with representative simple methods and GNNs across our proposed datasets show that LINKX achieves state-of-the-art performance for learning on non-homophilous graphs. Our codes and data are available at https://github.com/CUAI/Non-Homophily-Large-Scale.

Communication requires having a common language, a lingua franca, between agents. This language could emerge via a consensus process, but it may require many generations of trial and error. Alternatively, the lingua franca can be given by the environment, where agents ground their language in representations of the observed world. We demonstrate a simple way to ground language in learned representations, which facilitates decentralized multi-agent communication and coordination. We find that a standard representation learning algorithm -- autoencoding -- is sufficient for arriving at a grounded common language. When agents broadcast these representations, they learn to understand and respond to each other's utterances and achieve surprisingly strong task performance across a variety of multi-agent communication environments.

Automated data augmentation (ADA) techniques have played an important role in boosting the performance of deep models. Such techniques mostly aim to optimize a parameterized distribution over a discrete augmentation space. Thus, are restricted by the discretization of the search space which normally is handcrafted. To overcome the limitations, we take the first step to constructing a continuous mapping from $\mathbb{R}^d$ to image transformations (an augmentation space). Using this mapping, we take a novel approach where 1) we pose the ADA as a continuous optimization problem over the parameters of the augmentation distribution; and 2) use Stochastic Gradient Langevin Dynamics to learn and sample augmentations. This allows us to potentially explore the space of infinitely many possible augmentations, which otherwise was not possible due to the discretization of the space. This view of ADA is radically different from the standard discretization based view of ADA, and it opens avenues for utilizing the vast efficient gradient-based algorithms available for continuous optimization problems. Results over multiple benchmarks demonstrate the efficiency improvement of this work compared with previous methods.

Detecting manipulated images has become a significant emerging challenge. The advent of image sharing platforms and the easy availability of advanced photo editing software have resulted in a large quantities of manipulated images being shared on the internet. While the intent behind such manipulations varies widely, concerns on the spread of false news and misinformation is growing. Current state of the art methods for detecting these manipulated images suffers from the lack of training data due to the laborious labeling process. We address this problem in this paper, for which we introduce a manipulated image generation process that creates true positives using currently available datasets. Drawing from traditional work on image blending, we propose a novel generator for creating such examples. In addition, we also propose to further create examples that force the algorithm to focus on boundary artifacts during training. Strong experimental results validate our proposal.

In this paper, we propose a framework capable of generating face images that fall into the same distribution as that of a given one-shot example. We leverage a pre-trained StyleGAN model that already learned the generic face distribution. Given the one-shot target, we develop an iterative optimization scheme that rapidly adapts the weights of the model to shift the output's high-level distribution to the target's. To generate images of the same distribution, we introduce a style-mixing technique that transfers the low-level statistics from the target to faces randomly generated with the model. With that, we are able to generate an unlimited number of faces that inherit from the distribution of both generic human faces and the one-shot example. The newly generated faces can serve as augmented training data for other downstream tasks. Such setting is appealing as it requires labeling very few, or even one example, in the target domain, which is often the case of real-world face manipulations that result from a variety of unknown and unique distributions, each with extremely low prevalence. We show the effectiveness of our one-shot approach for detecting face manipulations and compare it with other few-shot domain adaptation methods qualitatively and quantitatively.

Recent advances in deep representation learning on Riemannian manifolds extend classical deep learning operations to better capture the geometry of the manifold. One possible extension is the Fr{é}chet mean, the generalization of the Euclidean mean; however, it has been difficult to apply because it lacks a closed form with an easily computable derivative. In this paper, we show how to differentiate through the Fr{é}chet mean for arbitrary Riemannian manifolds. Then, focusing on hyperbolic space, we derive explicit gradient expressions and a fast, accurate, and hyperparameter-free Fr{é}chet mean solver. This fully integrates the Fr{é}chet mean into the hyperbolic neural network pipeline. To demonstrate this integration, we present two case studies. First, we apply our Fr{é}chet mean to the existing Hyperbolic Graph Convolutional Network, replacing its projected aggregation to obtain state-of-the-art results on datasets with high hyperbolicity. Second, to demonstrate the Fr{é}chet mean’s capacity to generalize Euclidean neural network operations, we develop a hyperbolic batch normalization method that gives an improvement parallel to the one observed in the Euclidean setting.

Incorporating relational reasoning into neural networks has greatly expanded their capabilities and scope. One defining trait of relational reasoning is that it operates on a set of entities, as opposed to standard vector representations. Existing end-to-end approaches for relational reasoning typically extract entities from inputs by directly interpreting the latent feature representations as a set. We show that these approaches do not respect set permutational invariance and thus have fundamental representational limitations. To resolve this limitation, we propose a simple and general network module called Set Refiner Network (SRN). We first use synthetic image experiments to demonstrate how our approach effectively decomposes objects without explicit supervision. Then, we insert our module into existing relational reasoning models and show that respecting set invariance leads to substantial gains in prediction performance and robustness on several relational reasoning tasks. Code can be found at github.com/CUAI/BetterSetRepresentations.

To better conform to data geometry, recent deep generative modelling techniques adapt Euclidean constructions to non-Euclidean spaces. In this paper, we study normalizing flows on manifolds. Previous work has developed flow models for specific cases; however, these advancements hand craft layers on a manifold-by-manifold basis, restricting generality and inducing cumbersome design constraints. We overcome these issues by introducing Neural Manifold Ordinary Differential Equations, a manifold generalization of Neural ODEs, which enables the construction of Manifold Continuous Normalizing Flows (MCNFs). MCNFs require only local geometry (therefore generalizing to arbitrary manifolds) and compute probabilities with continuous change of variables (allowing for a simple and expressive flow construction). We find that leveraging continuous manifold dynamics produces a marked improvement for both density estimation and downstream tasks.

Neural networks are vulnerable to adversarial examples, malicious inputs crafted to fool trained models. Adversarial examples often exhibit black-box transfer, meaning that adversarial examples for one model can fool another model. However, adversarial examples are typically overfit to exploit the particular architecture and feature representation of a source model, resulting in sub-optimal black-box transfer attacks to other target models. We introduce the Intermediate Level Attack (ILA), which attempts to fine-tune an existing adversarial example for greater black-box transferability by increasing its perturbation on a pre-specified layer of the source model, improving upon state-of-the-art methods. We show that we can select a layer of the source model to perturb without any knowledge of the target models while achieving high transferability. Additionally, we provide some explanatory insights regarding our method and the effect of optimizing for adversarial examples using intermediate feature maps.

Recognizing objects from subcategories with very subtle differences remains a challenging task due to the large intra-class and small inter-class variation. Recent work tackles this problem in a weakly-supervised manner: object parts are first detected and the corresponding part-specific features are extracted for fine-grained classification. However, these methods typically treat the part-specific features of each image in isolation while neglecting their relationships between different images. In this paper, we propose Cross-X learning, a simple yet effective approach that exploits the relationships between different images and between different network layers for robust multi-scale feature learning. Our approach involves two novel components: (i) a cross-category cross-semantic regularizer that guides the extracted features to represent semantic parts and, (ii) a cross-layer regularizer that improves the robustness of multi-scale features by matching the prediction distribution across multiple layers. Our approach can be easily trained end-to-end and is scalable to large datasets like NABirds. We empirically analyze the contributions of different components of our approach and demonstrate its robustness, effectiveness and state-of-the-art performance on five benchmark datasets. Code is available at \url{https://github.com/cswluo/CrossX}.

Adversarial training has been shown to regularize deep neural networks in addition to increasing their robustness to adversarial examples. However, the regularization effect on very deep state of the art networks has not been fully investigated. In this paper, we present a novel approach to regularize deep neural networks by perturbing intermediate layer activations in an efficient manner. We use these perturbations to train very deep models such as ResNets and WideResNets and show improvement in performance across datasets of different sizes such as CIFAR-10, CIFAR-100 and ImageNet. Our ablative experiments show that the proposed approach not only provides stronger regularization compared to Dropout but also improves adversarial robustness comparable to traditional adversarial training approaches.

Visual Domain Adaptation is a problem of immense importance in computer vision. Previous approaches showcase the inability of even deep neural networks to learn informative representations across domain shift. This problem is more severe for tasks where acquiring hand labeled data is extremely hard and tedious. In this work, we focus on adapting the representations learned by segmentation networks across synthetic and real domains. Contrary to previous approaches that use a simple adversarial objective or superpixel information to aid the process, we propose an approach based on Generative Adversarial Networks (GANs) that brings the embeddings closer in the learned feature space. To showcase the generality and scalability of our approach, we show that we can achieve state of the art results on two challenging scenarios of synthetic to real domain adaptation. Additional exploratory experiments show that our approach: (1) generalizes to unseen domains and (2) results in improved alignment of source and target distributions.

How to best understand and interpret the decisions of deep neural networks is a crucial topic, as the impact of intelligent deep network systems is prevalent in many applications. We propose a superpixel based method to interpret and explain the results of black-box deep networks in the widely-applied image classification tasks. We perform probabilistic prediction difference analysis upon one or more superpixels clustered from image pixels. Our method generates a superpixel score map visualization that can provide rich interpretation regarding image components. Such interpretation provides supportive/unsupportive likelihood of image regions upon the decisions performed by the black-box classifier. We compare our method against state-of-art pixelwise interpretation methods over the latest deep neural network classifiers on the ImageNet dataset. Results show that our method produces more consistent interpretations in less computation time. Our method also supports interactive interpretation, where users can acquire explanations on specified regions through a convenient interface for a prompt reaction.

We present a Reinforcement Learning (RL) solution to the view planning problem (VPP), which generates a sequence of view points that are capable of sensing all accessible area of a given object represented as a 3D model. In doing so, the goal is to minimize the number of view points, making the VPP a class of set covering optimization problem (SCOP). The SCOP is NP-hard, and the inapproximability results tell us that the greedy algorithm provides the best approximation that runs in polynomial time. In order to find a solution that is better than the greedy algorithm, (i) we introduce a novel score function by exploiting the geometry of the 3D model, (ii) we device an intuitive approach to VPP using this score function, and (iii) we cast VPP as a Markovian Decision Process (MDP), and solve the MDP in RL framework using well-known RL algorithms. In particular, we use SARSA, Watkins-Q and TD with function approximation to solve the MDP. We compare the results of our method with the baseline greedy algorithm in an extensive set of test objects, and show that we can outperform the baseline in almost all cases.

In this work, we present the RNN Tree (RNN-T), an adaptive learning framework for skeleton based human action recognition. Our method categorizes action classes and uses multiple Recurrent Neural Networks (RNNs) in a treelike hierarchy. The RNNs in RNN-T are co-trained with the action category hierarchy, which determines the structure of RNN-T. Actions in skeletal representations are recognized via a hierarchical inference process, during which individual RNNs differentiate finer-grained action classes with increasing confidence. Inference in RNN-T ends when any RNN in the tree recognizes the action with high confidence, or a leaf node is reached. RNN-T effectively addresses two main challenges of large-scale action recognition: (i) able to distinguish fine-grained action classes that are intractable using a single network, and (ii) adaptive to new action classes by augmenting an existing model. We demonstrate the effectiveness of RNN-T/ACH method and compare it with the state-of-the-art methods on a large-scale dataset and several existing benchmarks.

Convolutional Neural Networks have been a subject of great importance over the past decade and great strides have been made in their utility for producing state of the art performance in many computer vision problems. However, the behavior of deep networks is yet to be fully understood and is still an active area of research. In this work, we present an intriguing behavior: pre-trained CNNs can be made to improve their predictions by structurally perturbing the input. We observe that these perturbations - referred as Guided Perturbations - enable a trained network to improve its prediction performance without any learning or change in network weights. We perform various ablative experiments to understand how these perturbations affect the local context and feature representations. Furthermore, we demonstrate that this idea can improve performance of several existing approaches on semantic segmentation and scene labeling tasks on the PASCAL VOC dataset and supervised classification tasks on MNIST and CIFAR10 datasets.

Drift is the most difficult issue in object visual tracking based on framework of “tracking-by-detection”. Due to the self-taught learning, the mis-aligned samples are potentially to be incorporated in learning and degrade the discrimination of the tracker. This paper proposes a new tracking approach that resolves this problem by three multi-level collaborative components: a high-level global appearance tracker provides a basic prediction, upon which the structure preserved low-level local patches matching helps to guarantee precise tracking with minimized drift. Those local patches are deliberately deployed on the foreground object via foreground/background segmentation, which is realized by a simple and efficient classifier trained by super-pixel segments. Experimental results show that the three closely collaborated components enable our tracker runs in real time and performs favourably against state-of-the-art approaches on challenging benchmark sequences.

In recent years, deep learning, particularly Convolutional Neural Network (CNN), has shown great efficacy for solving various vision tasks. In image segmentation, it has been demonstrated that a CNN can greatly outperform other approaches. However, special attention has to be paid towards setting various parameters in the CNN that affects the scale of the feature map generated at the last convolutional layer, where scale here refers to the ratio of the number of pixels in the original input image that correspond to each pixel in the feature map. Quite often, the optimal settings are tied to the specific problem on hand and can be fairly challenging to determine. To overcome such an issue, this paper proposes a multiscale Fully Convolutional Network (FCN) that combines networks trained at various scales, thereby allowing for conducting segmentation more generically. Moreover, such a multiscale architecture allows for incremental fine-tuning as more training images become available later on and new networks can be trained and added to the combined network. Such flexibility has great utility in applications such as industrial inspection, where training images may not be readily available initially, but yet requires a high level of accuracy. This paper will validate our findings by reporting the results that we have obtained by applying multiscale FCN to the inspection of aircraft engine part.

Rope shovels are widely used in the mining industry to dig ore. During operation, one or more teeth in the bucket can be lost as a result of the force that impacts the teeth, which causes a serious problem when the broken teeth gets picked up by the haul truck and eventually ended up in the crusher and jamming it. In this paper, we present a vision system for monitoring tooth condition and detecting missing tooth for a mining shovel. Our system leverages the predictable range of motion that the bucket of a rope shovel goes through during operation due to the camera mounting. For this reason, our strategy is to use exemplar based image retrieval and a sliding window procedure to first locate a pre-defined static region of the bucket in a live frame, followed by detecting the tooth line region based on its relative position to the selected exemplar. Once the tooth line region is detected, we proceed by determining its transformation from a recently detected tooth line region, and rectify the tooth line region before conducting image differencing. The difference image is then converted into a response map by sliding a tooth template associated with the retrieved exemplar and computing a correlation score per sliding window. Finally, the tooth line associated with the retrieved exemplar is rigidly scaled and rotated within a specified range in the response map to find a final tooth line position with the largest overall fitting score. An individual tooth is then flagged as missing if its missing measurement exceeds a threshold. The outstanding performance and high reliability of the proposed system have been demonstrated by experiments on video sequences collected from an iron ore mining site and a two-month trial of an installed unit on a production line.

We describe an algorithmic pipeline for automatic registration of smooth object image on 3D CAD model, which has many applications in industrial inspections. This enables camera pose estimation and 3D measurement of structures of interest from 2D images. In this work, when given a query image, we leverage image retrieval to find a similar historical image, in which its 2D points' corresponding locations on the 3D model has already been computed offline. By matching features between the query and retrieved image, the latter's feature locations on the 3D model can be transferred to the former. Camera pose of the query image can then be computed for overlaying it on the 3D model. A big challenge arises when the object of interest, such as one with smooth metallic surface, has extremely limited number of reliable features, which is especially true in industrial inspection. This causes popular recognition and detection algorithms as well as 3D calibration algorithms that depend on a sufficient number of good features to fail. We describe how we construct a pipeline to overcome this challenge. Comparative results and accuracy of 3D measurements made on real datasets with the proposed pipeline validate our approach.

Environments such as schools, public parks and prisons and others that contain a large number of people are typically characterized by frequent and complex social interactions. In order to identify activities and behaviors in such environments, it is necessary to understand the interactions that take place at a group level. To this end, this paper addresses the problem of detecting and predicting suspicious and in particular aggressive behaviors between groups of individuals such as gangs in prison yards. The work builds on a mature multi-camera multi-target person tracking system that operates in real-time and has the ability to handle crowded conditions. We consider two approaches for grouping individuals: (i) agglomerative clustering favored by the computer vision community, as well as (ii) decisive clustering based on the concept of modularity, which is favored by the social network analysis community. We show the utility of such grouping analysis towards the detection of group activities of interest. The presented algorithm is integrated with a system operating in real-time to successfully detect highly realistic aggressive behaviors enacted by correctional officers in a simulated prison environment. We present results from these enactments that demonstrate the efficacy of our approach.

This work addresses the important problem of the discovery and analysis of social networks from surveillance video. A computer vision approach to this problem is made possible by the proliferation of video data obtained from camera networks, particularly state-of-the-art Pan-Tilt-Zoom (PTZ) and tracking camera systems that have the capability to acquire high-resolution face images as well as tracks of people under challenging conditions. We perform “opportunistic” face recognition on captured images and compute motion similarities between tracks of people on the ground plane. To deal with the unknown correspondences between faces and tracks, we present a novel graph-cut based algorithm to solve this association problem. It enables the robust estimation of a social network that captures the interactions between individuals in spite of large amounts of noise in the datasets. We also introduce an algorithm that we call “modularity-cut”, which is an Eigen-analysis based approach for discovering community and leadership structure in the estimated social network. Our approach is illustrated with promising results from a fully integrated multi-camera system under challenging conditions over long period of time.

Background modeling and subtraction to detect new or moving objects in a scene is an important component of many intelligent video applications. Compared to a single camera, the use of multiple cameras leads to better handling of shadows, specularities and illumination changes due to the utilization of geometric information. Although the result of stereo matching can be used as the feature for detection, it has been shown that the detection process can be made much faster by a simple subtraction of the intensities observed at stereo-generated conjugate pairs in the two views. The methodology however, suffers from false and missed detections due to some geometric considerations. In this paper, we perform a detailed analysis of such errors. Then, we propose a sensor configuration that eliminates false detections. Algorithms are also proposed that effectively eliminate most detection errors due to missed detections, specular reflections and objects being geometrically close to the background. Experiments on several scenes illustrate the utility and enhanced performance of the proposed approach compared to existing techniques.

We describe a stereo rectification method suitable for automatic 3D surveillance. We take advantage of the fact that in a typical urban scene, there is ordinarily a small number of dominant planes. Given two views of the scene, we align a dominant plane in one view with the other. Conjugate epipolar lines between the reference view and plane-aligned image become geometrically identical and can be added to the rectified image pair line by line. Selecting conjugate epipolar lines to cover the whole image is simplified since they are geometrically identical. In addition, the polarities of conjugate epipolar lines are automatically preserved by plane alignment, which simplifies stereo matching.

We describe the design of a scalable and wide coverage visual surveillance system. Scalability (the ability to add and remove cameras easily during system operation with minimal overhead and system degradation) is achieved by utilizing only image-based information for camera control. We show that when a pan-tilt-zoom camera pans and tilts, a given image point moves in a circular and a linear trajectory, respectively. We create a scene model using a plan view of the scene. The scene model makes it easy for us to handle occlusion prediction and schedule video acquisition tasks subject to visibility constraints. We describe a maximum weight matching algorithm to assign cameras to tasks that meet the visibility constraints. The system is illustrated both through simulations and real video from a 6-camera configuration.

In automated surveillance systems with multiple cameras, the system must be able to position the cameras accurately. Each camera must be able to pan-tilt such that an object detected in the scene is in a vantage position in the camera's image plane and subsequently capture images of that object. Typically, camera calibration is required. We propose an approach that uses only image-based information. Each camera is assigned a pan-tilt zero-position. Position of an object detected in one camera is related to the other cameras by homographies between the zero-positions while different pan-tilt positions of the same camera are related in the form of projective rotations. We then derive that the trajectories in the image plane corresponding to these projective rotations are approximately circular for pan and linear for tilt. The camera control technique is subsequently tested in a working prototype.