CVPR 2024 Papers

We introduce "HallusionBench" a comprehensive benchmark designed for the evaluation of image-context reasoning. This benchmark presents significant challenges to advanced large visual-language models (LVLMs) such as GPT-4V(ision) Gemini Pro Vision Claude 3 and LLaVA-1.5 by emphasizing nuanced understanding and interpretation of visual data. The benchmark comprises 346 images paired with 1129 questions all meticulously crafted by human experts. We introduce a novel structure for these visual questions designed to establish control groups. This structure enables us to conduct a quantitative analysis of the models' response tendencies logical consistency and various failure modes. In our evaluation on HallusionBench we benchmarked 15 different models highlighting a 31.42% question-pair accuracy achieved by the state-of-the-art GPT-4V. Notably all other evaluated models achieve accuracy below 16%. Moreover our analysis not only highlights the observed failure modes including language hallucination and visual illusion but also deepens an under standing of these pitfalls. Our comprehensive case studies within HallusionBench shed light on the challenges of hallucination and illusion in LVLMs. Based on these insights we suggest potential pathways for their future improvement. The benchmark and codebase can be accessed at https://github.com/tianyilab/HallusionBench.

High-resolution wide-angle fisheye images are becoming more and more important for robotics applications such as autonomous driving. However using ordinary convolutional neural networks or vision transformers on this data is problematic due to projection and distortion losses introduced when projecting to a rectangular grid on the plane. We introduce the HEAL-SWIN transformer which combines the highly uniform Hierarchical Equal Area iso-Latitude Pixelation (HEALPix) grid used in astrophysics and cosmology with the Hierarchical Shifted-Window (SWIN) transformer to yield an efficient and flexible model capable of training on high-resolution distortion-free spherical data. In HEAL-SWIN the nested structure of the HEALPix grid is used to perform the patching and windowing operations of the SWIN transformer enabling the network to process spherical representations with minimal computational overhead. We demonstrate the superior performance of our model on both synthetic and real automotive datasets as well as a selection of other image datasets for semantic segmentation depth regression and classification tasks. Our code is publicly available.

Vision-language foundation models have shown remarkable performance in various zero-shot settings such as image retrieval classification or captioning. But so far those models seem to fall behind when it comes to zero-shot localization of referential expressions and objects in images. As a result they need to be fine-tuned for this task. In this paper we show that pretrained vision-language (VL) models allow for zero-shot open-vocabulary object localization without any fine-tuning. To leverage those capabilities we propose a Grounding Everything Module (GEM) that generalizes the idea of value-value attention introduced by CLIPSurgery to a self-self attention path. We show that the concept of self-self attention corresponds to clustering thus enforcing groups of tokens arising from the same object to be similar while preserving the alignment with the language space. To further guide the group formation we propose a set of regularizations that allows the model to finally generalize across datasets and backbones. We evaluate the proposed GEM framework on various benchmark tasks and datasets for semantic segmentation. GEM not only outperforms other training-free open-vocabulary localization methods but also achieves state-of-the-art results on the recently proposed OpenImagesV7 large-scale segmentation benchmark. Code is available at https://github.com/WalBouss/GEM

Recent advancements in Large Vision-Language Models (VLMs) have shown great promise in natural image domains allowing users to hold a dialogue about given visual content. However such general-domain VLMs perform poorly for Remote Sensing (RS) scenarios leading to inaccurate or fabricated information when presented with RS domain-specific queries. Such a behavior emerges due to the unique challenges introduced by RS imagery. For example to handle high-resolution RS imagery with diverse scale changes across categories and many small objects region-level reasoning is necessary alongside holistic scene interpretation. Furthermore the lack of domain-specific multimodal instruction following data as well as strong backbone models for RS make it hard for the models to align their behavior with user queries. To address these limitations we propose GeoChat - the first versatile remote sensing VLM that offers multitask conversational capabilities with high-resolution RS images. Specifically GeoChat can not only answer image-level queries but also accepts region inputs to hold region-specific dialogue. Furthermore it can visually ground objects in its responses by referring to their spatial coordinates. To address the lack of domain-specific datasets we generate a novel RS multimodal instruction-following dataset by extending image-text pairs from existing diverse RS datasets. Leveraging this rich dataset we fine-tune our remote sensing VLM based on the LLaVA-1.5 architecture. We establish a comprehensive benchmark for RS multitask conversations and compare with a number of baseline methods. GeoChat demonstrates robust zero-shot performance on various remote sensing tasks e.g. image and region captioning visual question answering scene classification visually grounded conversations and referring object detection. Our codes will be open-sourced.

Action understanding matters for intelligent agents and has attracted long-term attention. It can be formed as the mapping from the action physical space to the semantic space. Typically researchers built action datasets according to idiosyncratic choices to define classes and push the envelope of benchmarks respectively. Thus datasets are incompatible with each other like "Isolated Islands" due to semantic gaps and various class granularities e.g. do housework in dataset A and wash plate in dataset B. We argue that a more principled semantic space is an urgent need to concentrate the community efforts and enable us to use all datasets together to pursue generalizable action learning. To this end we design a structured action semantic space in view of verb taxonomy hierarchy and covering massive actions. By aligning the classes of previous datasets to our semantic space we gather (image/video/skeleton/MoCap) datasets into a unified database in a unified label system i.e. bridging "isolated islands" into a "Pangea". Accordingly we propose a novel model mapping from the physical space to semantic space to fully use Pangea. In extensive experiments our new system shows significant superiority especially in transfer learning. Our code and data will be made public at https://mvig-rhos.com/pangea.

Polarization is a fundamental property of light that encodes abundant information regarding surface shape material illumination and viewing geometry. The computer vision community has witnessed a blossom of polarization-based vision applications such as reflection removal shape-from-polarization (SfP) transparent object segmentation and color constancy partially due to the emergence of single-chip mono/color polarization sensors that make polarization data acquisition easier than ever. However is polarization-based vision vulnerable to adversarial attacks? If so is that possible to realize these adversarial attacks in the physical world without being perceived by human eyes? In this paper we warn the community of the vulnerability of polarization-based vision which can be more serious than RGB-based vision. By adapting a commercial LCD projector we achieve locally controllable polarizing projection which is successfully utilized to fool state-of-the-art polarization-based vision algorithms for glass segmentation and SfP. Compared with existing physical attacks on RGB-based vision which always suffer from the trade-off between attack efficacy and eye conceivability the adversarial attackers based on polarizing projection are contact-free and visually imperceptible since naked human eyes can rarely perceive the difference of viciously manipulated polarizing light and ordinary illumination. This poses unprecedented risks on polarization-based vision for which due attentions should be paid and counter measures be considered.

We introduce Florence-2 a novel vision foundation model with a unified prompt-based representation for various computer vision and vision-language tasks. While existing large vision models excel in transfer learning they struggle to perform diverse tasks with simple instructions a capability that implies handling the complexity of various spatial hierarchy and semantic granularity. Florence-2 was designed to take text-prompt as task instructions and generate desirable results in text forms whether it be captioning object detection grounding or segmentation. This multi-task learning setup demands large-scale high-quality annotated data. To this end we co-developed FLD-5B that consists of 5.4 billion comprehensive visual annotations on 126 million images using an iterative strategy of automated image annotation and model refinement. We adopted a sequence-to-sequence structure to train Florence-2 to perform versatile and comprehensive vision tasks. Extensive evaluations on numerous tasks demonstrated Florence-2 to be a strong vision foundation model contender with unprecedented zero-shot and fine-tuning capabilities.

Fine-grained action analysis in multi-person sports is complex due to athletes' quick movements and intense physical confrontations which result in severe visual obstructions in most scenes. In addition accessible multi-person sports video datasets lack fine-grained action annotations in both space and time adding to the difficulty in fine-grained action analysis. To this end we construct a new multi-person basketball sports video dataset named FineSports which contains fine-grained semantic and spatial-temporal annotations on 10000 NBA game videos covering 52 fine-grained action types 16000 action instances and 123000 spatial-temporal bounding boxes. We also propose a new prompt-driven spatial-temporal action location approach called PoSTAL composed of a prompt-driven target action encoder (PTA) and an action tube-specific detector (ATD) to directly generate target action tubes with fine-grained action types without any off-line proposal generation. Extensive experiments on the FineSports dataset demonstrate that PoSTAL outperforms state-of-the-art methods. Data and code are available at https://github.com/PKU-ICST-MIPL/FineSports_CVPR2024.

Existing action quality assessment (AQA) methods mainly learn deep representations at the video level for scoring diverse actions. Due to the lack of a fine-grained understanding of actions in videos they harshly suffer from low credibility and interpretability thus insufficient for stringent applications such as Olympic diving events. We argue that a fine-grained understanding of actions requires the model to perceive and parse actions in both time and space which is also the key to the credibility and interpretability of the AQA technique. Based on this insight we propose a new fine-grained spatial-temporal action parser named FineParser. It learns human-centric foreground action representations by focusing on target action regions within each frame and exploiting their fine-grained alignments in time and space to minimize the impact of invalid backgrounds during the assessment. In addition we construct fine-grained annotations of human-centric foreground action masks for the FineDiving dataset called FineDiving-HM. With refined annotations on diverse target action procedures FineDiving-HM can promote the development of real-world AQA systems. Through extensive experiments we demonstrate the effectiveness of FineParser which outperforms state-of-the-art methods while supporting more tasks of fine-grained action understanding. Data and code are available at https://github.com/PKU-ICST-MIPL/FineParser_CVPR2024.

Recent advances in neural network pruning have shown how it is possible to reduce the computational costs and memory demands of deep learning models before training. We focus on this framework and propose a new pruning at initialization algorithm that leverages the Neural Tangent Kernel (NTK) theory to align the training dynamics of the sparse network with that of the dense one. Specifically we show how the usually neglected data-dependent component in the NTK's spectrum can be taken into account by providing an analytical upper bound to the NTK's trace obtained by decomposing neural networks into individual paths. This leads to our Path eXclusion (PX) a foresight pruning method designed to preserve the parameters that mostly influence the NTK's trace. PX is able to find lottery tickets (i.e. good paths) even at high sparsity levels and largely reduces the need for additional training. When applied to pre-trained models it extracts subnetworks directly usable for several downstream tasks resulting in performance comparable to those of the dense counterpart but with substantial cost and computational savings.

Fairness is a critical concern in deep learning especially in healthcare where these models influence diagnoses and treatment decisions. Although fairness has been investigated in the vision-only domain the fairness of medical vision-language (VL) models remains unexplored due to the scarcity of medical VL datasets for studying fairness. To bridge this research gap we introduce the first fair vision-language medical dataset (Harvard-FairVLMed) that provides detailed demographic attributes ground-truth labels and clinical notes to facilitate an in-depth examination of fairness within VL foundation models. Using Harvard-FairVLMed we conduct a comprehensive fairness analysis of two widely-used VL models (CLIP and BLIP2) pre-trained on both natural and medical domains across four different protected attributes. Our results highlight significant biases in all VL models with Asian Male Non-Hispanic and Spanish being the preferred subgroups across the protected attributes of race gender ethnicity and language respectively. In order to alleviate these biases we propose FairCLIP an optimal-transport-based approach that achieves a favorable trade-off between performance and fairness by reducing the Sinkhorn distance between the overall sample distribution and the distributions corresponding to each demographic group. As the first VL dataset of its kind Harvard-FairVLMed holds the potential to catalyze advancements in the development of machine learning models that are both ethically aware and clinically effective. Our dataset and code are available at https://ophai.hms.harvard.edu/datasets/harvard-fairvlmed10k.

Despite noise and caption quality having been acknowledged as important factors impacting vision-language contrastive pre-training in this paper we show that the full potential of improving the training process by addressing such issues is yet to be realized. Specifically we firstly study and analyze two issues affecting training: incorrect assignment of negative pairs and low caption quality and diversity. Then we devise effective solutions for addressing both problems which essentially require training with multiple true positive pairs. Finally we propose training with sigmoid loss to address such a requirement. We show very large gains over the current state-of-the-art for both image recognition ( +6% on average over 11 datasets) and image retrieval ( +19% on Flickr30k and +15% on MSCOCO).

We study supervised action segmentation whose goal is to predict framewise action labels of a video. To capture temporal dependencies over long horizons prior works either improve framewise features with transformer or refine framewise predictions with learned action features. However they are computationally costly and ignore that frame and action features contain complimentary information which can be leveraged to enhance both features and improve temporal modeling. Therefore we propose an efficient Frame-Action Cross-attention Temporal modeling (FACT) framework that performs temporal modeling with frame and action features in parallel and leverage this parallelism to achieve iterative bidirectional information transfer between the features and refine them. FACT network contains (i) a frame branch to learn frame-level information with convolutions and frame features (ii) an action branch to learn action-level dependencies with transformers and action tokens and (iii) cross-attentions to allow communication between the two branches. We also propose a new matching loss to ensure each action token uniquely encodes an action segment thus better captures its semantics. Thanks to our architecture we can also leverage textual transcripts of videos to help action segmentation. We evaluate FACT on four video datasets (two egocentric and two third-person) for action segmentation with and without transcripts showing that it significantly improves the state-of-the-art accuracy while enjoys lower computational cost (3 times faster) than existing transformer-based methods

Large multi-modal models (LMMs) hold the potential to usher in a new era of automated visual assistance for people who are blind or low vision (BLV). Yet these models have not been systematically evaluated on data captured by BLV users. We address this by empirically assessing CLIP a widely-used LMM likely to underpin many assistive technologies. Testing 25 CLIP variants in a zero-shot classification task we find that their accuracy is 15 percentage points lower on average for images captured by BLV users than web-crawled images. This disparity stems from CLIP's sensitivities to 1) image content (e.g. not recognizing disability objects as well as other objects); 2) image quality (e.g. not being robust to lighting variation); and 3) text content (e.g. not recognizing objects described by tactile adjectives as well as visual ones). We delve deeper with a textual analysis of three common pre-training datasets: LAION-400M LAION-2B and DataComp-1B showing that disability content is rarely mentioned. We then provide three examples that illustrate how the performance disparities extend to three downstream models underpinned by CLIP: OWL-ViT CLIPSeg and DALL-E2. We find that few-shot learning with as few as 5 images can mitigate CLIP's quality-of-service disparities for BLV users in some scenarios which we discuss alongside a set of other possible mitigations.

Event cameras have recently been shown beneficial for practical vision tasks such as action recognition thanks to their high temporal resolution power efficiency and reduced privacy concerns. However current research is hindered by 1) the difficulty in processing events because of their prolonged duration and dynamic actions with complex and ambiguous semantics and 2) the redundant action depiction of the event frame representation with fixed stacks. We find language naturally conveys abundant semantic information rendering it stunningly superior in reducing semantic uncertainty. In light of this we propose ExACT a novel approach that for the first time tackles event-based action recognition from a cross-modal conceptualizing perspective. Our ExACT brings two technical contributions. Firstly we propose an adaptive fine-grained event (AFE) representation to adaptively filter out the repeated events for the stationary objects while preserving dynamic ones. This subtly enhances the performance of ExACT without extra computational cost. Then we propose a conceptual reasoning-based uncertainty estimation module which simulates the recognition process to enrich the semantic representation. In particular conceptual reasoning builds the temporal relation based on the action semantics and uncertainty estimation tackles the semantic uncertainty of actions based on the distributional representation. Experiments show that our ExACT achieves superior recognition accuracy of 94.83%(+2.23%) 90.10%(+37.47%) and 67.24% on PAF HARDVS and our SeAct datasets respectively.

We propose Strongly Supervised pre-training with ScreenShots (S4) - a novel pre-training paradigm for Vision-Language Models using data from large-scale web screenshot rendering. Using web screenshots unlocks a treasure trove of visual and textual cues that are not present in using image-text pairs. In S4 we leverage the inherent tree-structured hierarchy of HTML elements and the spatial localization to carefully design 10 pre-training tasks with large scale annotated data. These tasks resemble downstream tasks across different domains and the annotations are cheap to obtain. We demonstrate that compared to current screenshot pre-training objectives our innovative pre-training method significantly enhances performance of image-to-text model in nine varied and popular downstream tasks - up to 76.1% improvements on Table Detection and at least 1% on Widget Captioning.

Recently temporal action detection (TAD) has seen significant performance improvement with end-to-end training. However due to the memory bottleneck only models with limited scales and limited data volumes can afford end-to-end training which inevitably restricts TAD performance. In this paper we reduce the memory consumption for end-to-end training and manage to scale up the TAD backbone to 1 billion parameters and the input video to 1536 frames leading to significant detection performance. The key to our approach lies in our proposed temporal-informative adapter (TIA) which is a novel lightweight module that reduces training memory. Using TIA we free the humongous backbone from learning to adapt to the TAD task by only updating the parameters in TIA. TIA also leads to better TAD representation by temporally aggregating context from adjacent frames throughout the backbone. We evaluate our model across four representative datasets. Owing to our efficient design we are able to train end-to-end on VideoMAEv2-giant and achieve 75.4% mAP on THUMOS14 being the first end-to-end model to outperform the best feature-based methods.

The most performant spatio-temporal action localisation models use external person proposals and complex external memory banks. We propose a fully end-to-end transformer based model that directly ingests an input video and outputs tubelets -- a sequence of bounding boxes and the action classes at each frame. Our flexible model can be trained with either sparse bounding-box supervision on individual frames or full tubelet annotations. And in both cases it predicts coherent tubelets as the output. Moreover our end-to-end model requires no additional pre-processing in the form of proposals or post-processing in terms of non-maximal suppression. We perform extensive ablation experiments and significantly advance the state-of-the-art on five different spatio-temporal action localisation benchmarks with both sparse keyframes and full tubelet annotations.

Vision-language models (VLMs) have recently shown promising results in traditional downstream tasks. Evaluation studies have emerged to assess their abilities with the majority focusing on the third-person perspective and only a few addressing specific tasks from the first-person perspective. However the capability of VLMs to "think" from a first-person perspective a crucial attribute for advancing autonomous agents and robotics remains largely unexplored. To bridge this research gap we introduce EgoThink a novel visual question-answering benchmark that encompasses six core capabilities with twelve detailed dimensions. The benchmark is constructed using selected clips from egocentric videos with manually annotated question-answer pairs containing first-person information. To comprehensively assess VLMs we evaluate twenty-one popular VLMs on EgoThink. Moreover given the open-ended format of the answers we use GPT-4 as the automatic judge to compute single-answer grading. Experimental results indicate that although GPT-4V leads in numerous dimensions all evaluated VLMs still possess considerable potential for improvement in first-person perspective tasks. Meanwhile enlarging the number of trainable parameters has the most significant impact on model performance on EgoThink. In conclusion EgoThink serves as a valuable addition to existing evaluation benchmarks for VLMs providing an indispensable resource for future research in the realm of embodied artificial intelligence and robotics.