CVPR 2024 Papers

It is a long-lasting goal to design an embodied system that can solve long-horizon open-world tasks in human-like ways. However existing approaches usually struggle with compound difficulties caused by the logic-aware decomposition and context-aware execution of these tasks. To this end we introduce MP5 an open-ended multimodal embodied system built upon the challenging Minecraft simulator which can decompose feasible sub-objectives design sophisticated situation-aware plans and perform embodied action control with frequent communication with a goal-conditioned active perception scheme. Specifically MP5 is developed on top of recent advances in Multimodal Large Language Models (MLLMs) and the system is modulated into functional modules that can be scheduled and collaborated to ultimately solve pre-defined context- and process-dependent tasks. Extensive experiments prove that MP5 can achieve a 22% success rate on difficult process-dependent tasks and a 91% success rate on tasks that heavily depend on the context. Moreover MP5 exhibits a remarkable ability to address many open-ended tasks that are entirely novel.

Different from a unimodal model whose input is from a single modality the input (called multi-modal input) of a multi-modal model is from multiple modalities such as image 3D points audio text etc. Similar to unimodal models many existing studies show that a multi-modal model is also vulnerable to adversarial perturbation where an attacker could add small perturbation to all modalities of a multi-modal input such that the multi-modal model makes incorrect predictions for it. Existing certified defenses are mostly designed for unimodal models which achieve sub-optimal certified robustness guarantees when extended to multi-modal models as shown in our experimental results. In our work we propose MMCert the first certified defense against adversarial attacks to a multi-modal model. We derive a lower bound on the performance of our MMCert under arbitrary adversarial attacks with bounded perturbations to both modalities (e.g. in the context of auto-driving we bound the number of changed pixels in both RGB image and depth image). We evaluate our MMCert using two benchmark datasets: one for the multi-modal road segmentation task and the other for the multi-modal emotion recognition task. Moreover we compare our MMCert with a state-of-the-art certified defense extended from unimodal models. Our experimental results show that our MMCert outperforms the baseline.

Pre-trained Vision-Language Models (VLMs) have served as excellent foundation models for transfer learning in diverse downstream tasks. However tuning VLMs for few-shot generalization tasks faces a discrimination -- generalization dilemma i.e. general knowledge should be preserved and task-specific knowledge should be fine-tuned. How to precisely identify these two types of representations remains a challenge. In this paper we propose a Multi-Modal Adapter (MMA) for VLMs to improve the alignment between representations from text and vision branches. MMA aggregates features from different branches into a shared feature space so that gradients can be communicated across branches. To determine how to incorporate MMA we systematically analyze the discriminability and generalizability of features across diverse datasets in both the vision and language branches and find that (1) higher layers contain discriminable dataset-specific knowledge while lower layers contain more generalizable knowledge and (2) language features are more discriminable than visual features and there are large semantic gaps between the features of the two modalities especially in the lower layers. Therefore we only incorporate MMA to a few higher layers of transformers to achieve an optimal balance between discrimination and generalization. We evaluate the effectiveness of our approach on three tasks: generalization to novel classes novel target datasets and domain generalization. Compared to many state-of-the-art methods our MMA achieves leading performance in all evaluations. Code is at https://github.com/ZjjConan/Multi-Modal-Adapter

We present M&M VTO-a mix and match virtual try-on method that takes as input multiple garment images text description for garment layout and an image of a person. An example input includes: an image of a shirt an image of a pair of pants "rolled sleeves shirt tucked in" and an image of a person. The output is a visualization of how those garments (in the desired layout) would look like on the given person. Key contributions of our method are: 1) a single stage diffusion based model with no super resolution cascading that allows to mix and match multiple garments at 1024x512 resolution preserving and warping intricate garment details 2) architecture design (VTO UNet Diffusion Transformer) to disentangle denoising from person specific features allowing for a highly effective finetuning strategy for identity preservation (6MB model per individual vs 4GB achieved with e.g. dreambooth finetuning); solving a common identity loss problem in current virtual try-on methods 3) layout control for multiple garments via text inputs finetuned over PaLI-3 for virtual try-on task. Experimental results indicate that M&M VTO achieves state-of-the-art performance both qualitatively and quantitatively as well as opens up new opportunities for virtual try-on via language-guided and multi-garment try-on.

In recent years anchor-based methods have achieved promising progress in multi-view clustering. The performances of these methods are significantly affected by the quality of the anchors. However the anchors generated by previous works solely rely on single-view information ignoring the correlation among different views. In particular we observe that similar patterns are more likely to exist between similar views so such correlation information can be leveraged to enhance the quality of the anchors which is also omitted. To this end we propose a novel plug-and-play anchor enhancement strategy through view correlation for multi-view clustering. Specifically we construct a view graph based on aligned initial anchor graphs to explore inter-view correlations. By learning from view correlation we enhance the anchors of the current view using the relationships between anchors and samples on neighboring views thereby narrowing the spatial distribution of anchors on similar views. Experimental results on seven datasets demonstrate the superiority of our proposed method over other existing methods. Furthermore extensive comparative experiments validate the effectiveness of the proposed anchor enhancement module when applied to various anchor-based methods.

Multi-task learning has become increasingly popular in the machine learning field but its practicality is hindered by the need for large labeled datasets. Most multi-task learning methods depend on fully labeled datasets wherein each input example is accompanied by ground-truth labels for all target tasks. Unfortunately curating such datasets can be prohibitively expensive and impractical especially for dense prediction tasks which require per-pixel labels for each image. With this in mind we propose Joint-Task Regularization (JTR) an intuitive technique which leverages cross-task relations to simultaneously regularize all tasks in a single joint-task latent space to improve learning when data is not fully labeled for all tasks. JTR stands out from existing approaches in that it regularizes all tasks jointly rather than separately in pairs---therefore it achieves linear complexity relative to the number of tasks while previous methods scale quadratically. To demonstrate the validity of our approach we extensively benchmark our method across a wide variety of partially labeled scenarios based on NYU-v2 Cityscapes and Taskonomy.

Human intelligence can retrieve any person according to both visual and language descriptions. However the current computer vision community studies specific person re-identification (ReID) tasks in different scenarios separately which limits the applications in the real world. This paper strives to resolve this problem by proposing a new instruct-ReID task that requires the model to retrieve images according to the given image or language instructions. Our instruct-ReID is a more general ReID setting where existing 6 ReID tasks can be viewed as special cases by designing different instructions. We propose a large-scale OmniReID benchmark and an adaptive triplet loss as a baseline method to facilitate research in this new setting. Experimental results show that the proposed multi-purpose ReID model trained on our OmniReID benchmark without finetuning can improve +0.5% +0.6% +7.7% mAP on Market1501 MSMT17 CUHK03 for traditional ReID +6.4% +7.1% +11.2% mAP on PRCC VC-Clothes LTCC for clothes-changing ReID +11.7% mAP on COCAS+ real2 for clothes template based clothes-changing ReID when using only RGB images +24.9% mAP on COCAS+ real2 for our newly defined language-instructed ReID +4.3% on LLCM for visible-infrared ReID +2.6% on CUHK-PEDES for text-to-image ReID. The datasets the model and code are available at https://github.com/hwz-zju/Instruct-ReID.

Semantic segmentation in bird's eye view (BEV) plays a crucial role in autonomous driving. Previous methods usually follow an end-to-end pipeline directly predicting the BEV segmentation map from monocular RGB inputs. However the challenge arises when the RGB inputs and BEV targets from distinct perspectives making the direct point-to-point predicting hard to optimize. In this paper we decompose the original BEV segmentation task into two stages namely BEV map reconstruction and RGB-BEV feature alignment. In the first stage we train a BEV autoencoder to reconstruct the BEV segmentation maps given corrupted noisy latent representation which urges the decoder to learn fundamental knowledge of typical BEV patterns. The second stage involves mapping RGB input images into the BEV latent space of the first stage directly optimizing the correlations between the two views at the feature level. Our approach simplifies the complexity of combining perception and generation into distinct steps equipping the model to handle intricate and challenging scenes effectively. Besides we propose to transform the BEV segmentation map from the Cartesian to the polar coordinate system to establish the column-wise correspondence between RGB images and BEV maps. Moreover our method requires neither multi-scale features nor camera intrinsic parameters for depth estimation and saves computational overhead. Extensive experiments on nuScenes and Argoverse show the effectiveness and efficiency of our method. Code is available at https://github.com/happytianhao/TaDe.

The rise of multimodal large language models (MLLMs) has spurred interest in language-based driving tasks. However existing research typically focuses on limited tasks and often omits key multi-view and temporal information which is crucial for robust autonomous driving. To bridge these gaps we introduce NuInstruct a novel dataset with 91K multi-view video-QA pairs across 17 subtasks where each task demands holistic information (e.g. temporal multi-view and spatial) significantly elevating the challenge level. To obtain NuInstruct we propose a novel SQL-based method to generate instruction-response pairs automatically which is inspired by the driving logical progression of humans. We further present BEV-InMLLM an end-to-end method for efficiently deriving instruction-aware Bird's-Eye-View (BEV) features language-aligned for large language models. BEV-InMLLM integrates multi-view spatial awareness and temporal semantics to enhance MLLMs' capabilities on NuInstruct tasks. Moreover our proposed BEV injection module is a plug-and-play method for existing MLLMs. Our experiments on NuInstruct demonstrate that BEV-InMLLM significantly outperforms existing MLLMs e.g 9% improvement on various tasks. We release our NuInstruct at https://github.com/xmed-lab/NuInstruct.

This paper introduces Hierarchical Diffusion Policy (HDP) a hierarchical agent for multi-task robotic manipulation. HDP factorises a manipulation policy into a hierarchical structure: a high-level task-planning agent which predicts a distant next-best end-effector pose (NBP) and a low-level goal-conditioned diffusion policy which generates optimal motion trajectories. The factorised policy representation allows HDP to tackle both long-horizon task planning while generating fine-grained low-level actions. To generate context-aware motion trajectories while satisfying robot kinematics constraints we present a novel kinematics-aware goal-conditioned control agent Robot Kinematics Diffuser (RK-Diffuser). Specifically RK-Diffuser learns to generate both the end-effector pose and joint position trajectories and distill the accurate but kinematics-unaware end-effector pose diffuser to the kinematics-aware but less accurate joint position diffuser via differentiable kinematics. Empirically we show that HDP achieves a significantly higher success rate than the state-of-the-art methods in both simulation and real-world.

Multi-task visual scene understanding aims to leverage the relationships among a set of correlated tasks which are solved simultaneously by embedding them within a uni- fied network. However most existing methods give rise to two primary concerns from a task-level perspective: (1) the lack of task-independent correspondences for distinct tasks and (2) the neglect of explicit task-consensual dependencies among various tasks. To address these issues we propose a novel synergy embedding models (SEM) which goes be- yond multi-task dense prediction by leveraging two innova- tive designs: the intra-task hierarchy-adaptive module and the inter-task EM-interactive module. Specifically the con- structed intra-task module incorporates hierarchy-adaptive keys from multiple stages enabling the efficient learning of specialized visual patterns with an optimal trade-off. In ad- dition the developed inter-task module learns interactions from a compact set of mutual bases among various tasks benefiting from the expectation maximization (EM) algo- rithm. Extensive empirical evidence from two public bench- marks NYUD-v2 and PASCAL-Context demonstrates that SEM consistently outperforms state-of-the-art approaches across a range of metrics.

Point clouds captured by different sensors such as RGB-D cameras and LiDAR possess non-negligible domain gaps. Most existing methods design different network architectures and train separately on point clouds from various sensors. Typically point-based methods achieve outstanding performances on even-distributed dense point clouds from RGB-D cameras while voxel-based methods are more efficient for large-range sparse LiDAR point clouds. In this paper we propose geometry-to-voxel auxiliary learning to enable voxel representations to access point-level geometric information which supports better generalisation of the voxel-based backbone with additional interpretations of multi-sensor point clouds. Specifically we construct hierarchical geometry pools generated by a voxel-guided dynamic point network which efficiently provide auxiliary fine-grained geometric information adapted to different stages of voxel features. We conduct experiments on joint multi-sensor datasets to demonstrate the effectiveness of GeoAuxNet. Enjoying elaborate geometric information our method outperforms other models collectively trained on multi-sensor datasets and achieve competitive results with the-state-of-art experts on each single dataset.

In class incremental learning (CIL) scenarios the phenomenon of catastrophic forgetting caused by the classifier's bias towards the current task has long posed a significant challenge. It is mainly caused by the characteristic of discriminative models. With the growing popularity of the generative multi-modal models we would explore replacing discriminative models with generative ones for CIL. However transitioning from discriminative to generative models requires addressing two key challenges. The primary challenge lies in transferring the generated textual information into the classification of distinct categories. Additionally it requires formulating the task of CIL within a generative framework. To this end we propose a novel generative multi-modal model (GMM) framework for class incremental learning. Our approach directly generates labels for images using an adapted generative model. After obtaining the detailed text we use a text encoder to extract text features and employ feature matching to determine the most similar label as the classification prediction. In the conventional CIL settings we achieve significantly better results in long-sequence task scenarios. Under the Few-shot CIL setting we have improved by at least 14% over the current state-of-the-art methods with significantly less forgetting.

We present GALA a framework that takes as input a single-layer clothed 3D human mesh and decomposes it into complete multi-layered 3D assets. The outputs can then be combined with other assets to create novel clothed human avatars with any pose. Existing reconstruction approaches often treat clothed humans as a single-layer of geometry and overlook the inherent compositionality of humans with hairstyles clothing and accessories thereby limiting the utility of the meshes for down-stream applications. Decomposing a single-layer mesh into separate layers is a challenging task because it requires the synthesis of plausible geometry and texture for the severely occluded regions. Moreover even with successful decomposition meshes are not normalized in terms of poses and body shapes failing coherent composition with novel identities and poses. To address these challenges we propose to leverage the general knowledge of a pretrained 2D diffusion model as geometry and appearance prior for humans and other assets. We first separate the input mesh using the 3D surface segmentation extracted from multi-view 2D segmentations. Then we synthesize the missing geometry of different layers in both posed and canonical spaces using a novel pose-guided Score Distillation Sampling (SDS) loss. Once we complete inpainting high-fidelity 3D geometry we also apply the same SDS loss to its texture to obtain the complete appearance including the initially occluded regions. Through a series of decomposition steps we obtain multiple layers of 3D assets in a shared canonical space normalized in terms of poses and human shapes hence supporting effortless composition to novel identities and reanimation with novel poses. Our experiments demonstrate the effectiveness of our approach for decomposition canonicalization and composition tasks compared to existing solutions.

Video Motion Magnification (VMM) aims to reveal subtle and imperceptible motion information of objects in the macroscopic world. Prior methods directly model the motion field from the Eulerian perspective by Representation Learning that separates shape and texture or Multi-domain Learning from phase fluctuations. Inspired by the frequency spectrum we observe that the low-frequency components with stable energy always possess spatial structure and less noise making them suitable for modeling the subtle motion field. To this end we present FD4MM a new paradigm of Frequency Decoupling for Motion Magnification with a Multi-level Isomorphic Architecture to capture multi-level high-frequency details and a stable low-frequency structure (motion field) in video space. Since high-frequency details and subtle motions are susceptible to information degradation due to their inherent subtlety and unavoidable external interference from noise we carefully design Sparse High/Low-pass Filters to enhance the integrity of details and motion structures and a Sparse Frequency Mixer to promote seamless recoupling. Besides we innovatively design a contrastive regularization for this task to strengthen the model's ability to discriminate irrelevant features reducing undesired motion magnification. Extensive experiments on both Real-world and Synthetic Datasets show that our FD4MM outperforms SOTA methods. Meanwhile FD4MM reduces FLOPs by 1.63xand boosts inference speed by 1.68xthan the latest method. Our code is available at https://github.com/Jiafei127/FD4MM.

Federated Learning (FL) enables joint training across distributed clients using their local data privately. Federated Multi-Task Learning (FMTL) builds on FL to handle multiple tasks assuming model congruity that identical model architecture is deployed in each client. To relax this assumption and thus extend real-world applicability we introduce a novel problem setting Hetero-Client Federated Multi-Task Learning (HC-FMTL) to accommodate diverse task setups. The main challenge of HC-FMTL is the model incongruity issue that invalidates conventional aggregation methods. It also escalates the difficulties in model aggregation to deal with data and task heterogeneity inherent in FMTL. To address these challenges we propose the FedHCA^2 framework which allows for federated training of personalized models by modeling relationships among heterogeneous clients. Drawing on our theoretical insights into the difference between multi-task and federated optimization we propose the Hyper Conflict-Averse Aggregation scheme to mitigate conflicts during encoder updates. Additionally inspired by task interaction in MTL the Hyper Cross Attention Aggregation scheme uses layer-wise cross attention to enhance decoder interactions while alleviating model incongruity. Moreover we employ learnable Hyper Aggregation Weights for each client to customize personalized parameter updates. Extensive experiments demonstrate the superior performance of FedHCA^2 in various HC-FMTL scenarios compared to representative methods. Code is available at https://github.com/innovator-zero/FedHCA2.

In the realm of computer vision and robotics embodied agents are expected to explore their environment and carry out human instructions. This necessitates the ability to fully understand 3D scenes given their first-person observations and contextualize them into language for interaction. However traditional research focuses more on scene-level input and output setups from a global view. To address the gap we introduce EmbodiedScan a multi-modal ego-centric 3D perception dataset and benchmark for holistic 3D scene understanding. It encompasses over 5k scans encapsulating 1M ego-centric RGB-D views 1M language prompts 160k 3D-oriented boxes spanning over 760 categories some of which partially align with LVIS and dense semantic occupancy with 80 common categories. Building upon this database we introduce a baseline framework named Embodied Perceptron. It is capable of processing an arbitrary number of multi-modal inputs and demonstrates remarkable 3D perception capabilities both within the two series of benchmarks we set up i.e. fundamental 3D perception tasks and language-grounded tasks and in the wild.