CVPR 2024 Papers

The use of models that have been pre-trained on natural image datasets like ImageNet may face some limitations. First this use may be restricted due to copyright and license on the training images and privacy laws. Second these datasets and models may incorporate societal and ethical biases. Formula-driven supervised learning (FDSL) enables model pre-training to circumvent these issues. This consists of generating a synthetic image dataset based on mathematical formulae and pre-training the model on it. In this work we propose novel FDSL datasets based on Mandelbulb Variations. These datasets contain RGB images that are projections of colored objects deriving from the 3D Mandelbulb fractal. Pre-training ResNet-50 on one of our proposed datasets MandelbulbVAR-1k enables an average top-1 accuracy over target classification datasets that is at least 1% higher than pre-training on existing FDSL datasets. With regard to anomaly detection on MVTec AD pre-training the WideResNet-50 backbone on MandelbulbVAR-1k enables PatchCore to achieve 97.2% average image-level AUROC. This is only 1.9% lower than pre-training on ImageNet-1k (99.1%) and 4.5% higher than pre-training on the second-best performing FDSL dataset i.e. VisualAtom-1k (92.7%). Regarding Vision Transformer (ViT) pre-training another dataset that we propose and coin MandelbulbVAR-Hybrid-21k enables ViT-Base to achieve 82.2% top-1 accuracy on ImageNet-1k which is 0.4% higher than pre-training on ImageNet-21k (81.8%) and only 0.1% lower than pre-training on VisualAtom-1k (82.3%).

Few-Shot Class Incremental Learning (FSCIL) is a task that requires a model to learn new classes incrementally without forgetting when only a few samples for each class are given. FSCIL encounters two significant challenges: catastrophic forgetting and overfitting and these challenges have driven prior studies to primarily rely on shallow models such as ResNet-18. Even though their limited capacity can mitigate both forgetting and overfitting issues it leads to inadequate knowledge transfer during few-shot incremental sessions. In this paper we argue that large models such as vision and language transformers pre-trained on large datasets can be excellent few-shot incremental learners. To this end we propose a novel FSCIL framework called PriViLege Pre-trained Vision and Language transformers with prompting functions and knowledge distillation. Our framework effectively addresses the challenges of catastrophic forgetting and overfitting in large models through new pre-trained knowledge tuning (PKT) and two losses: entropy-based divergence loss and semantic knowledge distillation loss. Experimental results show that the proposed PriViLege significantly outperforms the existing state-of-the-art methods with a large margin e.g. +9.38% in CUB200 +20.58% in CIFAR-100 and +13.36% in miniImageNet. Our implementation code is available at https://github.com/KHU-AGI/PriViLege.

Domain Generalization (DG) aims to resolve distribution shifts between source and target domains and current DG methods are default to the setting that data from source and target domains share identical categories. Nevertheless there exists unseen classes from target domains in practical scenarios. To address this issue Open Set Domain Generalization (OSDG) has emerged and several methods have been exclusively proposed. However most existing methods adopt complex architectures with slight improvement compared with DG methods. Recently vision-language models (VLMs) have been introduced in DG following the fine-tuning paradigm but consume huge training overhead with large vision models. Therefore in this paper we innovate to transfer knowledge from VLMs to lightweight vision models and improve the robustness by introducing Perturbation Distillation (PD) from three perspectives including Score Class and Instance (SCI) named SCI-PD. Moreover previous methods are oriented by the benchmarks with identical and fixed splits ignoring the divergence between source domains. These methods are revealed to suffer from sharp performance decay with our proposed new benchmark Hybrid Domain Generalization (HDG) and a novel metric H^ 2 -CV which construct various splits to comprehensively assess the robustness of algorithms. Extensive experiments demonstrate that our method outperforms state-of-the-art algorithms on multiple datasets especially improving the robustness when confronting data scarcity.

Recent progress in Vision-Language (VL) foundation models has revealed the great advantages of cross-modality learning. However due to a large gap between vision and text they might not be able to sufficiently utilize the benefits of cross-modality information. In the field of human action recognition the additional pose modality may bridge the gap between vision and text to improve the effectiveness of cross-modality learning. In this paper we propose a novel framework called the Pose-enhanced Vision-Language (PeVL) model to adapt the VL model with pose modality to learn effective knowledge of fine-grained human actions. Our PeVL model includes two novel components: an Unsymmetrical Cross-Modality Refinement (UCMR) block and a Semantic-Guided Multi-level Contrastive (SGMC) module. The UCMR block includes Pose-guided Visual Refinement (P2V-R) and Visual-enriched Pose Refinement (V2P-R) for effective cross-modality learning. The SGMC module includes Multi-level Contrastive Associations of vision-text and pose-text at both action and sub-action levels and a Semantic-Guided Loss enabling effective contrastive learning with text. Built upon a pre-trained VL foundation model our model integrates trainable adapters and can be trained end-to-end. Our novel PeVL design over VL foundation model yields remarkable performance gains on four fine- grained human action recognition datasets achieving a new SOTA with a significantly small number of FLOPs for low- cost re-training.

This paper proposes a cross-modal distillation framework PartDistill which transfers 2D knowledge from vision-language models (VLMs) to facilitate 3D shape part segmentation. PartDistill addresses three major challenges in this task: the lack of 3D segmentation in invisible or undetected regions in the 2D projections inconsistent 2D predictions by VLMs and the lack of knowledge accumulation across different 3D shapes. PartDistill consists of a teacher network that uses a VLM to make 2D predictions and a student network that learns from the 2D predictions while extracting geometrical features from multiple 3D shapes to carry out 3D part segmentation. A bi-directional distillation including forward and backward distillations is carried out within the framework where the former forward distills the 2D predictions to the student network and the latter improves the quality of the 2D predictions which subsequently enhances the final 3D segmentation. Moreover PartDistill can exploit generative models that facilitate effortless 3D shape creation for generating knowledge sources to be distilled. Through extensive experiments PartDistill boosts the existing methods with substantial margins on widely used ShapeNetPart and PartNetE datasets by more than 15% and 12% higher mIoU scores respectively. The code for this work is available at https://github.com/ardianumam/PartDistill.

We address the challenge of content diversity and controllability in pedestrian simulation for driving scenarios. Recent pedestrian animation frameworks have a significant limitation wherein they primarily focus on either following trajectory or the content of the reference video consequently overlooking the potential diversity of human motion within such scenarios. This limitation restricts the ability to generate pedestrian behaviors that exhibit a wider range of variations and realistic motions and therefore restricts its usage to provide rich motion content for other components in the driving simulation system e.g. suddenly changed motion to which the autonomous vehicle should respond. In our approach we strive to surpass the limitation by showcasing diverse human motions obtained from various sources such as generated human motions in addition to following the given trajectory. The fundamental contribution of our framework lies in combining the motion tracking task with trajectory following which enables the tracking of specific motion parts (e.g. upper body) while simultaneously following the given trajectory by a single policy. This way we significantly enhance both the diversity of simulated human motion within the given scenario and the controllability of the content including language-based control. Our framework facilitates the generation of a wide range of human motions contributing to greater realism and adaptability in pedestrian simulations for driving scenarios.

Large pre-trained Vision-Language Models (VLMs) like CLIP despite having remarkable generalization ability are highly vulnerable to adversarial examples. This work studies the adversarial robustness of VLMs from the novel perspective of the text prompt instead of the extensively studied model weights (frozen in this work). We first show that the effectiveness of both adversarial attack and defense are sensitive to the used text prompt. Inspired by this we propose a method to improve resilience to adversarial attacks by learning a robust text prompt for VLMs. The proposed method named Adversarial Prompt Tuning (APT) is effective while being both computationally and data efficient. Extensive experiments are conducted across 15 datasets and 4 data sparsity schemes (from 1-shot to full training data settings) to show APT's superiority over hand-engineered prompts and other state-of-the-art adaption methods. APT demonstrated excellent abilities in terms of the in-distribution performance and the generalization under input distribution shift and across datasets. Surprisingly by simply adding one learned word to the prompts APT can significantly boost the accuracy and robustness (epsilon=4/255) over the hand-engineered prompts by +13% and +8.5% on average respectively. The improvement further increases in our most effective setting to +26.4% for accuracy and +16.7% for robustness. Code is available at https://github.com/TreeLLi/APT.

Recent Vision Transformer Compression (VTC) works mainly follow a two-stage scheme where the importance score of each model unit is first evaluated or preset in each submodule followed by the sparsity score evaluation according to the target sparsity constraint. Such a separate evaluation process induces the gap between importance and sparsity score distributions thus causing high search costs for VTC. In this work for the first time we investigate how to integrate the evaluations of importance and sparsity scores into a single stage searching the optimal subnets in an efficient manner. Specifically we present OFB a cost-efficient approach that simultaneously evaluates both importance and sparsity scores termed Once for Both (OFB) for VTC. First a bi-mask scheme is developed by entangling the importance score and the differentiable sparsity score to jointly determine the pruning potential (prunability) of each unit. Such a bi-mask search strategy is further used together with a proposed adaptive one-hot loss to realize the progressive-and-efficient search for the most important subnet. Finally Progressive Masked Image Modeling (PMIM) is proposed to regularize the feature space to be more representative during the search process which may be degraded by the dimension reduction. Extensive experiments demonstrate that OFB can achieve superior compression performance over state-of-the-art searching-based and pruning-based methods under various Vision Transformer architectures meanwhile promoting search efficiency significantly e.g. costing one GPU search day for the compression of DeiT-S on ImageNet-1K.

To interpret Vision Transformers post-hoc explanations assign salience scores to input pixels providing human-understandable heatmaps. However whether these interpretations reflect true rationales behind the model's output is still underexplored. To address this gap we study the faithfulness criterion of explanations: the assigned salience scores should represent the influence of the corresponding input pixels on the model's predictions. To evaluate faithfulness we introduce Salience-guided Faithfulness Coefficient (SaCo) a novel evaluation metric leveraging essential information of salience distribution. Specifically we conduct pair-wise comparisons among distinct pixel groups and then aggregate the differences in their salience scores resulting in a coefficient that indicates the explanation's degree of faithfulness. Our explorations reveal that current metrics struggle to differentiate between advanced explanation methods and Random Attribution thereby failing to capture the faithfulness property. In contrast our proposed SaCo offers a reliable faithfulness measurement establishing a robust metric for interpretations. Furthermore our SaCo demonstrates that the use of gradient and multi-layer aggregation can markedly enhance the faithfulness of attention-based explanation shedding light on potential paths for advancing Vision Transformer explainability.

We explore the boundaries of scaling up a multilingual vision and language model both in terms of size of the components and the breadth of its training task mixture. Our model achieves new levels of performance on a wide-range of varied and complex tasks including multiple image-based captioning and question-answering tasks image-based document understanding and few-shot (in-context) learning as well as object detection video question answering and video captioning. Our model advances the state-of-the-art on most vision-and-language benchmarks considered (20+ of them). Finally we observe emerging capabilities such as complex counting and multilingual object detection tasks that are not explicitly in the training mix.

Recent advances in Iterative Vision-and-Language Navigation(IVLN) introduce a more meaningful and practical paradigm of VLN by maintaining the agent's memory across tours of scenes. Although the long-term memory aligns better with the persistent nature of the VLN task it poses more challenges on how to utilize the highly unstructured navigation memory with extremely sparse supervision. Towards this end we propose OVER-NAV which aims to go over and beyond the current arts of IVLN techniques. In particular we propose to incorporate LLMs and open-vocabulary detectors to distill key information and establish correspondence between multi-modal signals. Such a mechanism introduces reliable cross-modal supervision and enables on-the-fly generalization to unseen scenes without the need of extra annotation and re-training. To fully exploit the interpreted navigation data we further introduce a structured representation coded Omnigraph to effectively integrate multi-modal information along the tour. Accompanied with a novel omnigraph fusion mechanism OVER-NAV is able to extract the most relevant knowledge from omnigraph for a more accurate navigating action. In addition OVER-NAV seamlessly supports both discrete and continuous environments under a unified framework. We demonstrate the superiority of OVER-NAV in extensive experiments.

Trajectory prediction is fundamental in computer vision and autonomous driving particularly for understanding pedestrian behavior and enabling proactive decision-making. Existing approaches in this field often assume precise and complete observational data neglecting the challenges associated with out-of-view objects and the noise inherent in sensor data due to limited camera range physical obstructions and the absence of ground truth for denoised sensor data. Such oversights are critical safety concerns as they can result in missing essential non-visible objects. To bridge this gap we present a novel method for out-of-sight trajectory prediction that leverages a vision-positioning technique. Our approach denoises noisy sensor observations in an unsupervised manner and precisely maps sensor-based trajectories of out-of-sight objects into visual trajectories. This method has demonstrated state-of-the-art performance in out-of-sight noisy sensor trajectory denoising and prediction on the Vi-Fi and JRDB datasets. By enhancing trajectory prediction accuracy and addressing the challenges of out-of-sight objects our work significantly contributes to improving the safety and reliability of autonomous driving in complex environments. Our work represents the first initiative towards Out-Of-Sight Trajectory prediction (OOSTraj) setting a new benchmark for future research.

Given the power of vision transformers a new learning paradigm pre-training and then prompting makes it more efficient and effective to address downstream visual recognition tasks. In this paper we identify a novel security threat towards such a paradigm from the perspective of backdoor attacks. Specifically an extra prompt token called the switch token in this work can turn the backdoor mode on i.e. converting a benign model into a backdoored one. Once under the backdoor mode a specific trigger can force the model to predict a target class. It poses a severe risk to the users of cloud API since the malicious behavior can not be activated and detected under the benign mode thus making the attack very stealthy. To attack a pre-trained model our proposed attack named SWARM learns a trigger and prompt tokens including a switch token. They are optimized with the clean loss which encourages the model always behaves normally even the trigger presents and the backdoor loss that ensures the backdoor can be activated by the trigger when the switch is on. Besides we utilize the cross-mode feature distillation to reduce the effect of the switch token on clean samples. The experiments on diverse visual recognition tasks confirm the success of our switchable backdoor attack i.e. achieving 95%+ attack success rate and also being hard to be detected and removed. Our code is available at https://github.com/20000yshust/SWARM.

Sequence-to-sequence vision-language models are showing promise but their applicability is limited by their inference latency due to their autoregressive way of generating predictions. We propose a parallel decoding sequence-to-sequence vision-language model trained with a Query-CTC loss that marginalizes over multiple inference paths in the decoder. This allows us to model the joint distribution of tokens rather than restricting to conditional distribution as in an autoregressive model. The resulting model NARVL achieves performance on-par with its state-of-the-art autoregressive counterpart but is faster at inference time reducing from the linear complexity associated with the sequential generation of tokens to a paradigm of constant time joint inference.

In this paper we investigate a new problem called narrative action evaluation (NAE). NAE aims to generate professional commentary that evaluates the execution of an action. Unlike traditional tasks such as score-based action quality assessment and video captioning involving superficial sentences NAE focuses on creating detailed narratives in natural language. These narratives provide intricate descriptions of actions along with objective evaluations. NAE is a more challenging task because it requires both narrative flexibility and evaluation rigor. One existing possible solution is to use multi-task learning where narrative language and evaluative information are predicted separately. However this approach results in reduced performance for individual tasks because of variations between tasks and differences in modality between language information and evaluation information. To address this we propose a prompt-guided multimodal interaction framework. This framework utilizes a pair of transformers to facilitate the interaction between different modalities of information. It also uses prompts to transform the score regression task into a video-text matching task thus enabling task interactivity. To support further research in this field we re-annotate the MTL-AQA and FineGym datasets with high-quality and comprehensive action narration. Additionally we establish benchmarks for NAE. Extensive experiment results prove that our method outperforms separate learning methods and naive multi-task learning methods. Data and code will be released at https://github.com/shiyi-zh0408/NAE_CVPR2024.

Action Localization is a challenging problem that combines detection and recognition tasks which are often addressed separately. State-of-the-art methods rely on off-the-shelf bounding box detections pre-computed at high resolution and propose transformer models that focus on the classification task alone. Such two-stage solutions are prohibitive for real-time deployment. On the other hand single-stage methods target both tasks by devoting part of the network (generally the backbone) to sharing the majority of the workload compromising performance for speed. These methods build on adding a DETR head with learnable queries that after cross- and self-attention can be sent to corresponding MLPs for detecting a person's bounding box and action. However DETR-like architectures are challenging to train and can incur in big complexity. In this paper we observe that a straight bipartite matching loss can be applied to the output tokens of a vision transformer. This results in a backbone + MLP architecture that can do both tasks without the need of an extra encoder-decoder head and learnable queries. We show that a single MViTv2-S architecture trained with bipartite matching to perform both tasks surpasses the same MViTv2-S when trained with RoI align on pre-computed bounding boxes. With a careful design of token pooling and the proposed training pipeline our Bipartite-Matching Vision Transformer model BMViT achieves +3 mAP on AVA2.2. w.r.t. the two-stage MViTv2-S counterpart. Code is available at https://github.com/IoannaNti/BMViT

While existing motion style transfer methods are effective between two motions with identical content their performance significantly diminishes when transferring style between motions with different contents. This challenge lies in the lack of clear separation between content and style of a motion. To tackle this challenge we propose a novel motion style transformer that effectively disentangles style from content and generates a plausible motion with transferred style from a source motion. Our distinctive approach to achieving the goal of disentanglement is twofold: (1) a new architecture for motion style transformer with 'part-attentive style modulator across body parts' and 'Siamese encoders that encode style and content features separately'; (2) style disentanglement loss. Our method outperforms existing methods and demonstrates exceptionally high quality particularly in motion pairs with different contents without the need for heuristic post-processing. Codes are available at https://github.com/Boeun-Kim/MoST.

Vision-language pre-trained models have achieved impressive performance on various downstream tasks. However their large model sizes hinder their utilization on platforms with limited computational resources. We find that directly using smaller pre-trained models and applying magnitude-based pruning on CLIP models leads to inflexibility and inferior performance. Recent efforts for VLP compression either adopt uni-modal compression metrics resulting in limited performance or involve costly mask-search processes with learnable masks. In this paper we first propose the Module-wise Pruning Error (MoPE) metric accurately assessing CLIP module importance by performance decline on cross-modal tasks. Using the MoPE metric we introduce a unified pruning framework applicable to both pre-training and task-specific fine-tuning compression stages. For pre-training MoPE-CLIP effectively leverages knowledge from the teacher model significantly reducing pre-training costs while maintaining strong zero-shot capabilities. For fine-tuning consecutive pruning from width to depth yields highly competitive task-specific models. Extensive experiments in two stages demonstrate the effectiveness of the MoPE metric and MoPE-CLIP outperforms previous state-of-the-art VLP compression methods.