We consider indoor 3D object detection with respect to a single RGB(-D) frame acquired from a commodity handheld device. We seek to significantly advance the status quo with respect to both data and modeling. First, we establish that existing datasets have significant limitations to scale, accuracy, and diversity of objects. As a result, we introduce the Cubify-Anything 1M (CA-1M) dataset, which exhaustively labels over 400K 3D objects on over 1K highly accurate laser-scanned scenes with near-perfect registration to over 3.5K handheld, egocentric captures. Next, we establish Cubify Transformer (CuTR), a fully Transformer 3D object detection baseline which rather than operating in 3D on point or voxel-based representations, predicts 3D boxes directly from 2D features derived from RGB(-D) inputs. While this approach lacks any 3D inductive biases, we show that paired with CA-1M, CuTR outperforms pointbased methods — accurately recalling over 62%c of objects in 3D, and is significantly more capable at handling noise and uncertainty present in commodity LiDAR-derived depth maps while also providing promising RGB only performance without architecture changes. Furthermore, by pre-training on CA-1M, CuTR can outperform point-based methods on a more diverse variant of SUN RGB-D — supporting the notion that while inductive biases in 3D are useful at the smaller sizes of existing datasets, they fail to scale to the data-rich regime of CA-1M. Overall, this dataset and baseline model provide strong evidence that we are moving towards models which can effectively Cubify Anything.

We introduce FlexTok, a tokenizer that projects 2D images into variable-length, ordered 1D token sequences. For example, a 256x256 image can be resampled into anywhere from 1 to 256 discrete tokens, hierarchically and semantically compressing its information. By training a rectified flow model as the decoder and using nested dropout, FlexTok produces plausible reconstructions regardless of the chosen token sequence length. We evaluate our approach in an autoregressive generation setting using a simple GPT-style Transformer. On ImageNet, this approach achieves an FID$<$2 across 8 to 128 tokens, outperforming TiTok and matching state-of-the-art methods with far fewer tokens. We further extend the model to support to text-conditioned image generation and examine how FlexTok relates to traditional 2D tokenization. A key finding is that FlexTok enables next-token prediction to describe images in a coarse-to-fine "visual vocabulary", and that the number of tokens to generate depends on the complexity of the generation task.

We revisit scene-level 3D object detection as the output of an object-centric framework capable of both localization and mapping using 3D oriented boxes as the underlying geometric primitive. While existing 3D object detection approaches operate globally and implicitly rely on the a priori existence of metric camera poses, our method, Rooms from Motion (RfM) operates on a collection of un-posed images. By replacing the standard 2D keypoint-based matcher of structure-from-motion with an object-centric matcher based on image-derived 3D boxes, we estimate metric camera poses, object tracks, and finally produce a global, semantic 3D object map. When a priori pose is available, we can significantly improve map quality through optimization of global 3D boxes against individual observations. RfM shows strong localization performance and subsequently produces maps of higher quality than leading point-based and multi-view 3D object detection methods on CA-1M and ScanNet++, despite these global methods relying on overparameterization through point clouds or dense volumes. Rooms from Motion achieves a general, object-centric representation which not only extends the work of Cubify Anything to full scenes but also allows for inherently sparse localization and parametric mapping proportional to the number of objects in a scene.

We introduce UniGen, a unified multimodal large language model (MLLM) capable of image understanding and generation. We study the full training pipeline of UniGen from a data-centric perspective, including multi-stage pre-training, supervised fine-tuning, and direct preference optimization. More importantly, we propose a new Chain-of-Thought Verification (CoT-V) strategy for test-time scaling, which significantly boosts UniGen’s image generation quality using a simple Best-of-N test-time strategy. Specifically, CoT-V enables UniGen to act as both image generator and verifier at test time, assessing the semantic alignment between a text prompt and its generated image in a step-by-step CoT manner. Trained entirely on open-source datasets across all stages, UniGen achieves state-of-the-art performance on a range of image understanding and generation benchmarks, with a final score of 0.78 on GenEval and 85.19 on DPG-Bench. Through extensive ablation studies, our work provides actionable insights and addresses key challenges in the full life cycle of building unified MLLMs, contributing meaningful directions to future research. Code is available at https://github.com/apple/ml-unigen.

We present StreamBridge, a simple yet effective framework that seamlessly transforms offline Video-LLMs into streaming-capable models. It addresses two fundamental challenges in adapting existing models into online scenarios: (1) limited capability for multi-turn real-time understanding, and (2) lack of proactive response mechanisms. Specifically, StreamBridge incorporates (1) a memory buffer combined with a round-decayed compression strategy, supporting long-context multi-turn interactions, and (2) a decoupled, lightweight activation model that can be effortlessly integrated into existing Video-LLMs, enabling continuous proactive responses. To further support StreamBridge, we construct Stream-IT, a large-scale dataset tailored for streaming video understanding, featuring interleaved video-text sequences and diverse instruction formats. Extensive experiments show that StreamBridge significantly improves the streaming understanding capabilities of offline Video-LLMs across various tasks, outperforming even proprietary models such as GPT-4o and Gemini 1.5 Pro. Simultaneously, it achieves competitive or superior performance on standard video understanding benchmarks.

Current multimodal and multitask foundation models, like 4M or UnifiedIO, show promising results. However, their out-of-the-box abilities to accept diverse inputs and perform diverse tasks are limited by the (usually small) number of modalities and tasks they are trained on. In this paper, we develop a single any-to-any model trained on tens of highly diverse modalities and by performing co-training on large-scale multimodal datasets and text corpora. This includes training on images and text along with several semantic and geometric modalities, feature maps from recent state of the art models like DINOv2 and ImageBind, pseudo labels of specialist models like SAM and 4DHumans, and a range of new modalities that allow for novel ways to interact with the model and steer the generation, for example, image metadata or color palettes. A crucial step in this process is performing discrete tokenization on various modalities, whether they are image-like, neural network feature maps, vectors, structured data like instance segmentation or human poses, or data that can be represented as text. Through this, we show the possibility of training one model to solve at least 3x more tasks/modalities than existing models and doing so without a loss in performance. In addition, this enables more fine-grained and controllable multimodal generation capabilities and allows studying the distillation of models trained on diverse data and objectives into one unified model. We scale the training to a three billion parameter and different datasets. The multimodal models and training code are open sourced at https://4m.epfl.ch/.

Current machine learning models for vision are often highly specialized and limited to a single modality and task. In contrast, recent large language models exhibit a wide range of capabilities, hinting at a possibility for similarly versatile models in computer vision. In this paper, we take a step in this direction and propose a multimodal training scheme called 4M. It consists of training a single unified Transformer encoder-decoder using a masked modeling objective across a wide range of input/output modalities – including text, images, geometric, and semantic modalities, as well as neural network feature maps. 4M achieves scalability by unifying the representation space of all modalities through mapping them into discrete tokens and performing multimodal masked modeling on a small randomized subset of tokens. 4M leads to models that exhibit several key capabilities: (1) they can perform a diverse set of vision tasks out of the box, (2) they excel when fine-tuned for unseen downstream tasks or new input modalities, and (3) they can function as a generative model that can be conditioned on arbitrary modalities, enabling a wide variety of expressive multimodal editing capabilities with remarkable flexibility. Through experimental analyses, we demonstrate the potential of 4M for training versatile and scalable foundation models for vision tasks, setting the stage for further exploration in multimodal learning for vision and other domains.

We introduce GAUDI, a generative model capable of capturing the distribution of complex and realistic 3D scenes that can be rendered immersively from a moving camera. We tackle this challenging problem with a scalable yet powerful approach, where we first optimize a latent representation that disentangles radiance fields and camera poses. This latent representation is then used to learn a generative model that enables both unconditional and conditional generation of 3D scenes. Our model generalizes previous works that focus on single objects by removing the assumption that the camera pose distribution can be shared across samples. We show that GAUDI obtains state-of-the-art performance in the unconditional generative setting across multiple datasets and allows for conditional generation of 3D scenes given conditioning variables like sparse image observations or text that describes the scene.

In this work, we propose a tracker that differs from most existing multi-target trackers in two major ways. First, our tracker does not rely on a pre-trained object detector to get the initial object hypotheses. Second, our tracker's final output is the fine contours of the targets rather than traditional bounding boxes. Therefore, our tracker simultaneously solves three main problems: detection, data association and segmentation. This is especially important because the output of each of those three problems are highly correlated and the solution of one can greatly help improve the others. The proposed algorithm consists of two main components: structured learning and Lagrange dual decomposition. Our structured learning based tracker learns a model for each target and infers the best locations of all targets simultaneously in a video clip. The inference of our structured learning is achieved through a new Target Identity-aware Network Flow (TINF), where each node in the network encodes the probability of each target identity belonging to that node. The probabilities are obtained by training target specific models using a global structured learning technique. This is followed by proposed Lagrangian relaxation optimization to find the high quality solution to the network. This forms the first component of our tracker. The second component is Lagrange dual decomposition, which combines the structured learning tracker with a segmentation algorithm. For segmentation, multi-label Conditional Random Field (CRF) is applied to a superpixel based spatio-temporal graph in a segment of video, in order to assign background or target labels to every superpixel. We show how the multi-label CRF is combined with the structured learning tracker through our dual decomposition formulation. This leads to more accurate segmentation results and also helps better resolve typical difficulties in multiple target tracking, such as occlusion handling, ID-switch and track drifting. The experiments on diverse and challenging sequences show that our method achieves superior results compared to competitive approaches for detection, multiple target tracking as well as segmentation.

Multi-object tracking has been studied for decades. However, when it comes to tracking pedestrians in extremely crowded scenes, we are limited to only few works. This is an important problem which gives rise to several challenges. Pre-trained object detectors fail to localize targets in crowded sequences. This consequently limits the use of data-association based multi-target tracking methods which rely on the outcome of an object detector. Additionally, the small apparent target size makes it challenging to extract features to discriminate targets from their surroundings. Finally, the large number of targets greatly increases computational complexity which in turn makes it hard to extend existing multi-target tracking approaches to high-density crowd scenarios. In this paper, we propose a tracker that addresses the aforementioned problems and is capable of tracking hundreds of people efficiently. We formulate online crowd tracking as Binary Quadratic Programing. Our formulation employs target's individual information in the form of appearance and motion as well as contextual cues in the form of neighborhood motion, spatial proximity and grouping, and solves detection and data association simultaneously. In order to solve the proposed quadratic optimization efficiently, where state-of art commercial quadratic programing solvers fail to find the solution in a reasonable amount of time, we propose to use the most recent version of the Modified Frank Wolfe algorithm, which takes advantage of SWAP-steps to speed up the optimization. We show that the proposed formulation can track hundreds of targets efficiently and improves state-of-art results by significant margins on eleven challenging high density crowd sequences.

Data association is the backbone to many multiple object tracking (MOT) methods. In this paper we formulate data association as a Generalized Maximum Multi Clique problem (GMMCP). We show that this is the ideal case of modeling tracking in real world scenario where all the pairwise relationships between targets in a batch of frames are taken into account. Previous works assume simplified version of our tracker either in problem formulation or problem optimization. However, we propose a solution using GMMCP where no simplification is assumed in either steps. We show that the NP hard problem of GMMCP can be formulated through Binary-Integer Program where for small and medium size MOT problems the solution can be found efficiently. We further propose a speed-up method, employing Aggregated Dummy Nodes for modeling occlusion and miss-detection, which reduces the size of the input graph without using any heuristics. We show that, using the speedup method, our tracker lends itself to real-time implementation which is plausible in many applications. We evaluated our tracker on six challenging sequences of Town Center, TUD-Crossing, TUD-Stadtmitte, Parking-lot 1, Parking-lot 2 and Parking-lot pizza and show favorable improvement against state of art.

In this paper we show that multiple object tracking (MOT) can be formulated in a framework, where the detection and data-association are performed simultaneously. Our method allows us to overcome the confinements of data association based MOT approaches; where the performance is dependent on the object detection results provided at input level. At the core of our method lies structured learning which learns a model for each target and infers the best location of all targets simultaneously in a video clip. The inference of our structured learning is done through a new Target Identity-aware Network Flow (TINF), where each node in the network encodes the probability of each target identity belonging to that node. The proposed Lagrangian relaxation optimization finds the high quality solution to the network. During optimization a soft spatial constraint is enforced between the nodes of the graph which helps reducing the ambiguity caused by nearby targets with similar appearance in crowded scenarios. We show that automatically detecting and tracking targets in a single framework can help resolve the ambiguities due to frequent occlusion and heavy articulation of targets. Our experiments involve challenging yet distinct datasets and show that our method can achieve results better than the state-of-art.

A video captures a sequence and interactions of concepts that can be static, for instance, objects or scenes, or dynamic, such as actions. For large datasets containing hundreds of thousands of images or videos, it is impractical to manually annotate all the concepts, or all the instances of a single concept. However, a dictionary with visually-distinct elements can be created automatically from unlabeled videos which can capture and express the entire dataset. The downside to this machine-discovered dictionary is meaninglessness, i.e., its elements are devoid of semantics and interpretation. In this paper, we present an approach that leverages the strengths of semantic concepts and the machine-discovered DOVE by learning a relationship between them. Since instances of a semantic concept share visual similarity, the proposed approach uses soft-consensus regularization to learn the mapping that enforces instances from each semantic concept to have similar representations. The testing is performed by projecting the query onto the DOVE as well as new representations of semantic concepts from training, with non-negativity and unit summation constraints for probabilistic interpretation. We tested our formulation on TRECVID MED and SIN tasks, and obtained encouraging results.

Recent years have seen a major push for face recognition technology due to the large expansion of image sharing on social networks. In this paper, we consider the difficult task of determining parent-offspring resemblance using deep learning to answer the question "Who do I look like?" Although humans can perform this job at a rate higher than chance, it is not clear how they do it [2]. However, recent studies in anthropology [24] have determined which features tend to be the most discriminative. In this study, we aim to not only create an accurate system for resemblance detection, but bridge the gap between studies in anthropology with computer vision techniques. Further, we aim to answer two key questions: 1) Do offspring resemble their parents? and 2) Do offspring resemble one parent more than the other? We propose an algorithm that fuses the features and metrics discovered via gated autoencoders with a discriminative neural network layer that learns the optimal, or what we call genetic, features to delineate parent-offspring relationships. We further analyze the correlation between our automatically detected features and those found in anthropological studies. Meanwhile, our method outperforms the state-of-the-art in kinship verification by 3-10% depending on the relationship using specific (father-son, mother-daughter, etc.) and generic models.

Complex event recognition is an expanding research area aiming to recognize entities of high-level semantics in videos. Typical approaches exploit the so-called “bags” of spatiotemporal features such as STIP, ISA and DTF-HOG; yet, more recently, the notion of concept has emerged as an alternative, intermediate representation with greater descriptive power, and “bags of concepts” have been used for recognition. In this paper we argue that concepts in an event tend to articulate over a discernible temporal structure and we exploit a temporal model using the scores of concept detectors as measurements. In addition, we propose several heuristics to improve the initialization of the model's latent states and take advantage of the time-sparsity of the concepts. Experimental results on videos from the challenging TRECVID MED 2012 dataset show that the proposed approach achieves an improvement in average precision of 8.92% over comparable bags of concepts, thus validating the use of temporal structure over concepts for complex event recognition.

There is a large variety of trackers, which have been proposed in the literature during the last two decades with some mixed success. Object tracking in realistic scenarios is a difficult problem, therefore, it remains a most active area of research in computer vision. A good tracker should perform well in a large number of videos involving illumination changes, occlusion, clutter, camera motion, low contrast, specularities, and at least six more aspects. However, the performance of proposed trackers have been evaluated typically on less than ten videos, or on the special purpose datasets. In this paper, we aim to evaluate trackers systematically and experimentally on 315 video fragments covering above aspects. We selected a set of nineteen trackers to include a wide variety of algorithms often cited in literature, supplemented with trackers appearing in 2010 and 2011 for which the code was publicly available. We demonstrate that trackers can be evaluated objectively by survival curves, Kaplan Meier statistics, and Grubs testing. We find that in the evaluation practice the F-score is as effective as the object tracking accuracy (OTA) score. The analysis under a large variety of circumstances provides objective insight into the strengths and weaknesses of trackers.

We propose an approach to improve the detection performance of a generic detector when it is applied to a particular video. The performance of offline-trained objects detectors are usually degraded in unconstrained video environments due to variant illuminations, backgrounds and camera viewpoints. Moreover, most object detectors are trained using Haar-like features or gradient features but ignore video specific features like consistent color patterns. In our approach, we apply a Super pixel-based Bag-of-Words (BoW) model to iteratively refine the output of a generic detector. Compared to other related work, our method builds a video-specific detector using super pixels, hence it can handle the problem of appearance variation. Most importantly, using Conditional Random Field (CRF) along with our super pixel-based BoW model, we develop and algorithm to segment the object from the background. Therefore our method generates an output of the exact object regions instead of the bounding boxes generated by most detectors. In general, our method takes detection bounding boxes of a generic detector as input and generates the detection output with higher average precision and precise object regions. The experiments on four recent datasets demonstrate the effectiveness of our approach and significantly improves the state-of-art detector by 5-16% in average precision.

In this paper we investigate a new problem called visual business recognition. Automatic identification of businesses in images is an interesting task with plenty of potential applications especially for mobile device users. We propose a multimodal approach which incorporates business directories, textual information, and web images in a unified framework. We assume the query image is associated with a coarse location tag and utilize business directories for extracting an over complete list of nearby businesses which may be visible in the image. We use the name of nearby businesses as search keywords in order to automatically collect a set of relevant images from the web and perform image matching between them and the query. Additionally, we employ a text processing method customized for business recognition which is assisted by nearby business names; we fuse the information acquired from image matching and text processing in a probabilistic framework to recognize the businesses. We tested the proposed algorithm on a challenging set of user-uploaded and street view images with promising results for this new application.

Single camera-based multiple-person tracking is often hindered by difficulties such as occlusion and changes in appearance. In this paper, we address such problems by proposing a robust part-based tracking-by-detection framework. Human detection using part models has become quite popular, yet its extension in tracking has not been fully explored. Our approach learns part-based person-specific SVM classifiers which capture the articulations of the human bodies in dynamically changing appearance and background. With the part-based model, our approach is able to handle partial occlusions in both the detection and the tracking stages. In the detection stage, we select the subset of parts which maximizes the probability of detection, which significantly improves the detection performance in crowded scenes. In the tracking stage, we dynamically handle occlusions by distributing the score of the learned person classifier among its corresponding parts, which allows us to detect and predict partial occlusions, and prevent the performance of the classifiers from being degraded. Extensive experiments using the proposed method on several challenging sequences demonstrate state-of-the-art performance in multiple-people tracking.