CVPR 2024 Papers

Integrating whole-slide images (WSIs) and bulk transcriptomics for predicting patient survival can improve our understanding of patient prognosis. However this multimodal task is particularly challenging due to the different nature of these data: WSIs represent a very high-dimensional spatial description of a tumor while bulk transcriptomics represent a global description of gene expression levels within that tumor. In this context our work aims to address two key challenges: (1) how can we tokenize transcriptomics in a semantically meaningful and interpretable way? and (2) how can we capture dense multimodal interactions between these two modalities? Here we propose to learn biological pathway tokens from transcriptomics that can encode specific cellular functions. Together with histology patch tokens that encode the slide morphology we argue that they form appropriate reasoning units for interpretability. We fuse both modalities using a memory-efficient multimodal Transformer that can model interactions between pathway and histology patch tokens. Our model SURVPATH achieves state-of-the-art performance when evaluated against unimodal and multimodal baselines on five datasets from The Cancer Genome Atlas. Our interpretability framework identifies key multimodal prognostic factors and as such can provide valuable insights into the interaction between genotype and phenotype. Code available at https://github.com/mahmoodlab/SurvPath.

From content moderation to wildlife conservation the number of applications that require models to recognize nuanced or subjective visual concepts is growing. Traditionally developing classifiers for such concepts requires substantial manual effort measured in hours days or even months to identify and annotate data needed for training. Even with recently proposed Agile Modeling techniques which enable rapid bootstrapping of image classifiers users are still required to spend 30 minutes or more of monotonous repetitive data labeling just to train a single classifier. Drawing on Fiske's Cognitive Miser theory we propose a new framework that alleviates manual effort by replacing human labeling with natural language interactions reducing the total effort required to define a concept by an order of magnitude: from labeling 2000 images to only 100 plus some natural language interactions. Our framework leverages recent advances in foundation models both large language models and vision-language models to carve out the concept space through conversation and by automatically labeling training data points. Most importantly our framework eliminates the need for crowd-sourced annotations. Moreover our framework ultimately produces lightweight classification models that are deployable in cost-sensitive scenarios. Across 15 subjective concepts and across 2 public image classification datasets our trained models outperform traditional Agile Modeling as well as state-of-the-art zero-shot classification models like ALIGN CLIP CuPL and large visual question answering models like PaLI-X.

Model Inversion (MI) attacks aim to reconstruct private training data by abusing access to machine learning models. Contemporary MI attacks have achieved impressive attack performance posing serious threats to privacy. Meanwhile all existing MI defense methods rely on regularization that is in direct conflict with the training objective resulting in noticeable degradation in model utility. In this work we take a different perspective and propose a novel and simple Transfer Learning-based Defense against Model Inversion (TL-DMI) to render MI-robust models. Particularly by leveraging TL we limit the number of layers encoding sensitive information from private training dataset thereby degrading the performance of MI attack. We conduct an analysis using Fisher Information to justify our method. Our defense is remarkably simple to implement. Without bells and whistles we show in extensive experiments that TL-DMI achieves state-of-the-art (SOTA) MI robustness. Our code pre-trained models demo and inverted data are available at: https://hosytuyen.github.io/projects/TL-DMI

When adopting a deep learning model for embodied agents it is required that the model structure be optimized for specific tasks and operational conditions. Such optimization can be static such as model compression or dynamic such as adaptive inference. Yet these techniques have not been fully investigated for embodied control systems subject to time constraints which necessitate sequential decision-making for multiple tasks each with distinct inference latency limitations. In this paper we present MoDeC a time constraint-aware embodied control framework using the modular model adaptation. We formulate model adaptation to varying operational conditions on resource and time restrictions as dynamic routing on a modular network incorporating these conditions as part of multi-task objectives. Our evaluation across several vision-based embodied environments demonstrates the robustness of MoDeC showing that it outperforms other model adaptation methods in both performance and adherence to time constraints in robotic manipulation and autonomous driving applications.

Generalizability in deep neural networks plays a pivotal role in medical image segmentation. However deep learning-based medical image analyses tend to overlook the importance of frequency variance which is critical element for achieving a model that is both modality-agnostic and domain-generalizable. Additionally various models fail to account for the potential information loss that can arise from multi-task learning under deep supervision a factor that can impair the model's representation ability. To address these challenges we propose a Modality-agnostic Domain Generalizable Network (MADGNet) for medical image segmentation which comprises two key components: a Multi-Frequency in Multi-Scale Attention (MFMSA) block and Ensemble Sub-Decoding Module (E-SDM). The MFMSA block refines the process of spatial feature extraction particularly in capturing boundary features by incorporating multi-frequency and multi-scale features thereby offering informative cues for tissue outline and anatomical structures. Moreover we propose E-SDM to mitigate information loss in multi-task learning with deep supervision especially during substantial upsampling from low resolution. We evaluate the segmentation performance of MADGNet across six modalities and fifteen datasets. Through extensive experiments we demonstrate that MADGNet consistently outperforms state-of-the-art models across various modalities showcasing superior segmentation performance. This affirms MADGNet as a robust solution for medical image segmentation that excels in diverse imaging scenarios. Our MADGNet code is available in GitHub Link.

Video-based Unsupervised Domain Adaptation (VUDA) method improves the generalization of the video model enabling it to be applied to action recognition tasks in different environments. However these methods require continuous access to source data during the adaptation process which are impractical in real scenarios where the source videos are not available with concerns in transmission efficiency or privacy issues. To address this problem in this paper we propose to solve the Multimodal Video Test-Time Adaptation task (MVTTA). Existing image-based TTA methods cannot be directly applied to this task because video have domain shift in multimodal and temporal which brings difficulties to adaptation. To address the above challenges we propose a Modality-Collaborative Test-Time Adaptation (MC-TTA) Network. We maintain teacher and student memory banks respectively for generating pseudo-prototypes and target-prototypes. In the teacher model we propose Self-assembled Source-friendly Feature Reconstruction (SSFR) module to encourage the teacher memory bank to store features that are more likely to be consistent with the source distribution. Through multimodal prototype alignment and cross-modal relative consistency our method can effectively alleviate domain shift in videos. We evaluate the proposed model on four public video datasets. The results show that our model outperforms existing state-of-the-art methods.

Humans possess the capability to comprehend diverse modalities and seamlessly transfer information between them. In this work we introduce ModaVerse a Multi-modal Large Language Model (MLLM) capable of comprehending and transforming content across various modalities including images videos and audio. Predominant MLLM frameworks have largely relied on aligning latent spaces of textual and non-textual features. This alignment process which synchronizes a language model trained on textual data with encoders and decoders trained on multi-modal data often necessitates extensive training of several projection layers in multiple stages. Inspired by LLM-as-agent methodologies we propose a novel Input/Output (I/O) alignment mechanism that operates directly at the level of natural language. It aligns the LLM's output with the input of generative models avoiding the complexities associated with latent feature alignments and simplifying the multiple training stages of existing MLLMs into a single efficient process. By conducting experiments on several benchmarks we demonstrate that our approach attains comparable performance with the state of the art while achieving considerable efficiencies in data usage.

We present a lightweight and affordable motion capture method based on two smartwatches and a head-mounted camera. In contrast to the existing approaches that use six or more expert-level IMU devices our approach is much more cost-effective and convenient. Our method can make wearable motion capture accessible to everyone everywhere enabling 3D full-body motion capture in diverse environments. As a key idea to overcome the extreme sparsity and ambiguities of sensor inputs with different modalities we integrate 6D head poses obtained from the head-mounted cameras for motion estimation. To enable capture in expansive indoor and outdoor scenes we propose an algorithm to track and update floor level changes to define head poses coupled with a multi-stage Transformer-based regression module. We also introduce novel strategies leveraging visual cues of egocentric images to further enhance the motion capture quality while reducing ambiguities. We demonstrate the performance of our method on various challenging scenarios including complex outdoor environments and everyday motions including object interactions and social interactions among multiple individuals.

Contrastive pre-training of image-text foundation models such as CLIP demonstrated excellent zero-shot performance and improved robustness on a wide range of downstream tasks. However these models utilize large transformer-based encoders with significant memory and latency overhead which pose challenges for deployment on mobile devices. In this work we introduce MobileCLIP - a new family of efficient image-text models optimized for runtime performance along with a novel and efficient training approach namely multi-modal reinforced training. The proposed training approach leverages knowledge transfer from an image captioning model and an ensemble of strong CLIP encoders to improve the accuracy of efficient models. Our approach avoids train-time compute overhead by storing the additional knowledge in a reinforced dataset. MobileCLIP sets a new state-of-the-art latency-accuracy tradeoff for zero-shot classification and retrieval tasks on several datasets. Our MobileCLIP-S2 variant is 2.3x faster while more accurate compared to previous best CLIP model based on ViT-B/16. We further demonstrate the effectiveness of our multi-modal reinforced training by training a CLIP model based on ViT-B/16 image backbone and achieving +2.9% average performance improvement on 38 evaluation benchmarks compared to the previous best. Moreover we show that the proposed approach achieves 10x-1000x improved learning efficiency when compared with non- reinforced CLIP training. Code and models are available at https://github.com/apple/ml-mobileclip

Many existing motion prediction approaches rely on symbolic perception outputs to generate agent trajectories such as bounding boxes road graph information and traffic lights. This symbolic representation is a high-level abstraction of the real world which may render the motion prediction model vulnerable to perception errors (e.g. failures in detecting open-vocabulary obstacles) while missing salient information from the scene context (e.g. poor road conditions). An alternative paradigm is end-to-end learning from raw sensors. However this approach suffers from the lack of interpretability and requires significantly more training resources. In this work we propose tokenizing the visual world into a compact set of scene elements and then leveraging pre-trained image foundation models and LiDAR neural networks to encode all the scene elements in an open-vocabulary manner. The image foundation model enables our scene tokens to encode the general knowledge of the open world while the LiDAR neural network encodes geometry information. Our proposed representation can efficiently encode the multi-frame multi-modality observations with a few hundred tokens and is compatible with most transformer-based architectures. To evaluate our method we have augmented Waymo Open Motion Dataset with camera embeddings. Experiments over Waymo Open Motion Dataset show that our approach leads to significant performance improvements over the state-of-the-art.

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.

Reconstructing an avatar from a portrait image has many applications in multimedia but remains a challenging research problem. Extracting reflectance maps and geometry from one image is ill-posed: recovering geometry is a one-to-many mapping problem and reflectance and light are difficult to disentangle. Accurate geometry and reflectance can be captured under the controlled conditions of a light stage but it is costly to acquire large datasets in this fashion. Moreover training solely with this type of data leads to poor generalization with in-the-wild images. This motivates the introduction of MoSAR a method for 3D avatar generation from monocular images. We propose a semi-supervised training scheme that improves generalization by learning from both light stage and in-the-wild datasets. This is achieved using a novel differentiable shading formulation. We show that our approach effectively disentangles the intrinsic face parameters producing relightable avatars. As a result MoSAR estimates a richer set of skin reflectance maps and generates more realistic avatars than existing state-of-the-art methods. We also release a new dataset that provides intrinsic face attributes (diffuse specular ambient occlusion and translucency maps) for 10k subjects.

This paper addresses the task of video question answering (videoQA) via a decomposed multi-stage modular reasoning framework. Previous modular methods have shown promise with a single planning stage ungrounded in visual content. However through a simple and effective baseline we find that such systems can lead to brittle behavior in practice for challenging videoQA settings. Thus unlike traditional single-stage planning methods we propose a multi-stage system consisting of an event parser a grounding stage and a final reasoning stage in conjunction with an external memory. All stages are training-free and performed using few-shot prompting of large models creating interpretable intermediate outputs at each stage. By decomposing the underlying planning and task complexity our method MoReVQA improves over prior work on standard videoQA benchmarks (NExT-QA iVQA EgoSchema and ActivityNet-QA) with state-of-the-art results and extensions to related tasks (grounded videoQA paragraph captioning).

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.

We introduce MoMask a novel masked modeling framework for text-driven 3D human motion generation. In MoMask a hierarchical quantization scheme is employed to represent human motion as multi-layer discrete motion tokens with high-fidelity details. Starting at the base layer with a sequence of motion tokens obtained by vector quantization the residual tokens of increasing orders are derived and stored at the subsequent layers of the hierarchy. This is consequently followed by two distinct bidirectional transformers. For the base-layer motion tokens a Masked Transformer is designated to predict randomly masked motion tokens conditioned on text input at training stage. During generation (i.e. inference) stage starting from an empty sequence our Masked Transformer iteratively fills up the missing tokens; Subsequently a Residual Transformer learns to progressively predict the next-layer tokens based on the results from current layer. Extensive experiments demonstrate that MoMask outperforms the state-of-art methods on the text-to-motion generation task with an FID of 0.045 (vs e.g. 0.141 of T2M-GPT) on the HumanML3D dataset and 0.228 (vs 0.514) on KIT-ML respectively. MoMask can also be seamlessly applied in related tasks without further model fine-tuning such as text-guided temporal inpainting.

In the academic field the research on human motion prediction tasks mainly focuses on exploiting the observed information to forecast human movements accurately in the near future horizon. However a significant gap appears when it comes to the application field as current models are all trained offline with fixed parameters that are inherently suboptimal to handle the complex yet ever-changing nature of human behaviors. To bridge this gap in this paper we introduce the task of online meta adaptation for human motion prediction based on the insight that finding "smart weights" capable of swift adjustments to suit different motion contexts along the time is a key to improving predictive accuracy. We propose MoML which ingeniously borrows the bilevel optimization spirit of model-agnostic meta-learning to transform previous predictive mistakes into strong inductive biases to guide online adaptation. This is achieved by our MoAdapter blocks that can learn error information by facilitating efficient adaptation via a few gradient steps which fine-tunes our meta-learned "smart" initialization produced by the generic predictor. Considering real-time requirements in practice we further propose Fast-MoML a more efficient variant of MoML that features a closed-form solution instead of conventional gradient update. Experimental results show that our approach can effectively bring many existing offline motion prediction models online and improves their predictive accuracy.

The success of contrastive language-image pretraining (CLIP) relies on the supervision from the pairing between images and captions which tends to be noisy in web-crawled data. We present Mixture of Data Experts (MoDE) and learn a system of CLIP data experts via clustering. Each data expert is trained on one data cluster being less sensitive to false negative noises in other clusters. At inference time we ensemble their outputs by applying weights determined through the correlation between task metadata and cluster conditions. To estimate the correlation precisely the samples in one cluster should be semantically similar but the number of data experts should still be reasonable for training and inference. As such we consider the ontology in human language and propose to use fine-grained cluster centers to represent each data expert at a coarse-grained level. Experimental studies show that four CLIP data experts on ViT-B/16 outperform the ViT-L/14 by OpenAI CLIP and OpenCLIP on zero-shot image classification but with less (<35%) training cost. Meanwhile MoDE can train all data expert asynchronously and can flexibly include new data experts. The code is available here.

Learning-based stereo matching techniques have made significant progress. However existing methods inevitably lose geometrical structure information during the feature channel generation process resulting in edge detail mismatches. In this paper the Motif Channel Attention Stereo Matching Network (MoCha-Stereo) is designed to address this problem. We provide the Motif Channel Correlation Volume (MCCV) to determine more accurate edge matching costs. MCCV is achieved by projecting motif channels which capture common geometric structures in feature channels onto feature maps and cost volumes. In addition edge variations in the reconstruction error map also affect details matching we propose the Reconstruction Error Motif Penalty (REMP) module to further refine the full-resolution disparity estimation. REMP integrates the frequency information of typical channel features from the reconstruction error. MoCha-Stereo ranks 1st on the KITTI-2015 and KITTI-2012 Reflective leaderboards. Our structure also shows excellent performance in Multi-View Stereo. Code is avaliable at https://github.com/ZYangChen/MoCha-Stereo.

While federated learning (FL) systems often utilize quantization to battle communication and computational bottlenecks they have heretofore been limited to deploying fixed-precision quantization schemes. Meanwhile the concept of mixed-precision quantization (MPQ) where different layers of a deep learning model are assigned varying bit-width remains unexplored in the FL settings. We present a novel FL algorithm FedMPQ which introduces mixed-precision quantization to resource-heterogeneous FL systems. Specifically local models quantized so as to satisfy bit-width constraint are trained by optimizing an objective function that includes a regularization term which promotes reduction of precision in some of the layers without significant performance degradation. The server collects local model updates de-quantizes them into full-precision models and then aggregates them into a global model. To initialize the next round of local training the server relies on the information learned in the previous training round to customize bit-width assignments of the models delivered to different clients. In extensive benchmarking experiments on several model architectures and different datasets in both iid and non-iid settings FedMPQ outperformed the baseline FL schemes that utilize fixed-precision quantization while incurring only a minor computational overhead on the participating devices.