CVPR 2024 Papers

Model stealing (MS) involves querying and observing the output of a machine learning model to steal its capabilities. The quality of queried data is crucial yet obtaining a large amount of real data for MS is often challenging. Recent works have reduced reliance on real data by using generative models. However when high-dimensional query data is required these methods are impractical due to the high costs of querying and the risk of model collapse. In this work we propose using sample gradients (SG) to enhance the utility of each real sample as SG provides crucial guidance on the decision boundaries of the victim model. However utilizing SG in the model stealing scenario faces two challenges: 1. Pixel-level gradient estimation requires extensive query volume and is susceptible to defenses. 2. The estimation of sample gradients has a significant variance. This paper proposes Superpixel Sample Gradient stealing (SPSG) for model stealing under the constraint of limited real samples. With the basic idea of imitating the victim model's low-variance patch-level gradients instead of pixel-level gradients SPSG achieves efficient sample gradient estimation through two steps. First we perform patch-wise perturbations on query images to estimate the average gradient in different regions of the image. Then we filter the gradients through a threshold strategy to reduce variance. Exhaustive experiments demonstrate that with the same number of real samples SPSG achieves accuracy agreements and adversarial success rate significantly surpassing the current state-of-the-art MS methods. Codes are available at https://github.com/zyl123456aB/SPSG_attack.

Estimating the 3D structure of the human body from nat- ural scenes is a fundamental aspect of visual perception. 3D human pose estimation is a vital step in advancing fields like AIGC and human-robot interaction serving as a crucial tech- nique for understanding and interacting with human actions in real-world settings. However the current datasets often collected under single laboratory conditions using complex motion capture equipment and unvarying backgrounds are insufficient. The absence of datasets on variable conditions is stalling the progress of this crucial task. To facilitate the development of 3D pose estimation we present FreeMan the first large-scale multi-view dataset collected under the real- world conditions. FreeMan was captured by synchronizing 8 smartphones across diverse scenarios. It comprises 11M frames from 8000 sequences viewed from different perspec- tives. These sequences cover 40 subjects across 10 different scenarios each with varying lighting conditions. We have also established an semi-automated pipeline containing er- ror detection to reduce the workload of manual check and ensure precise annotation. We provide comprehensive eval- uation baselines for a range of tasks underlining the sig- nificant challenges posed by FreeMan. Further evaluations of standard indoor/outdoor human sensing datasets reveal that FreeMan offers robust representation transferability in real and complex scenes. FreeMan is publicly available at https://wangjiongw.github.io/freeman.

Photometric stereo is a well-established technique to estimate the surface normal of an object. However the requirement of capturing multiple high dynamic range images under different illumination conditions limits the speed and real-time applications. This paper introduces EventPS a novel approach to real-time photometric stereo using an event camera. Capitalizing on the exceptional temporal resolution dynamic range and low bandwidth characteristics of event cameras EventPS estimates surface normal only from the radiance changes significantly enhancing data efficiency. EventPS seamlessly integrates with both optimization-based and deep-learning-based photometric stereo techniques to offer a robust solution for non-Lambertian surfaces. Extensive experiments validate the effectiveness and efficiency of EventPS compared to frame-based counterparts. Our algorithm runs at over 30 fps in real-world scenarios unleashing the potential of EventPS in time-sensitive and high-speed downstream applications.

Being able to map the activities of others into one's own point of view is one fundamental human skill even from a very early age. Taking a step toward understanding this human ability we introduce EgoExoLearn a large-scale dataset that emulates the human demonstration following process in which individuals record egocentric videos as they execute tasks guided by demonstration videos. Focusing on the potential applications of daily assistance and professional support EgoExoLearn contains egocentric and demonstration video data spanning 120 hours captured in daily life scenarios and specialized laboratories. Along with the videos we record high-quality gaze data and provide detailed multimodal annotations formulating a playground for modeling the human ability to bridge asynchronous procedural actions from different viewpoints. To this end we present benchmarks such as cross-view association cross-view action planning and cross-view referenced skill assessment along with detailed analysis. We expect EgoExoLearn can serve as an important resource for bridging the actions across views thus paving the way for creating AI agents capable of seamlessly learning by observing humans in the real world. The dataset and benchmark codes are available at https://github.com/OpenGVLab/EgoExoLearn.

Domain shift is a formidable issue in Machine Learning that causes a model to suffer from performance degradation when tested on unseen domains. Federated Domain Generalization (FedDG) attempts to train a global model using collaborative clients in a privacy-preserving manner that can generalize well to unseen clients possibly with domain shift. However most existing FedDG methods either cause additional privacy risks of data leakage or induce significant costs in client communication and computation which are major concerns in the Federated Learning paradigm. To circumvent these challenges here we introduce a novel architectural method for FedDG namely gPerXAN which relies on a normalization scheme working with a guiding regularizer. In particular we carefully design Personalized eXplicitly Assembled Normalization to enforce client models selectively filtering domain-specific features that are biased towards local data while retaining discrimination of those features. Then we incorporate a simple yet effective regularizer to guide these models in directly capturing domain-invariant representations that the global model's classifier can leverage. Extensive experimental results on two benchmark datasets i.e. PACS and Office-Home and a real-world medical dataset Camelyon17 indicate that our proposed method outperforms other existing methods in addressing this particular problem.

This paper presents DriveTrack a new benchmark and data generation framework for long-range keypoint tracking in real-world videos. DriveTrack is motivated by the observation that the accuracy of state-of-the-art trackers depends strongly on visual attributes around the selected keypoints such as texture and lighting. The problem is that these artifacts are especially pronounced in real-world videos but these trackers are unable to train on such scenes due to a dearth of annotations. DriveTrack bridges this gap by building a framework to automatically annotate point tracks on autonomous driving datasets. We release a dataset consisting of 1 billion point tracks across 24 hours of video which is seven orders of magnitude greater than prior real-world benchmarks and on par with the scale of synthetic benchmarks. DriveTrack unlocks new use cases for point tracking in real-world videos. First we show that fine-tuning keypoint trackers on DriveTrack improves accuracy on real-world scenes by up to 7%. Second we analyze the sensitivity of trackers to visual artifacts in real scenes and motivate the idea of running assistive keypoint selectors alongside trackers.

Motion capture from a limited number of body-worn sensors such as inertial measurement units (IMUs) and pressure insoles has important applications in health human performance and entertainment. Recent work has focused on accurately reconstructing whole-body motion from a specific sensor configuration using six IMUs. While a common goal across applications is to use the minimal number of sensors to achieve required accuracy the optimal arrangement of the sensors might differ from application to application. We propose a single diffusion model DiffusionPoser which reconstructs human motion in real-time from an arbitrary combination of sensors including IMUs placed at specified locations and pressure insoles. Unlike existing methods our model grants users the flexibility to determine the number and arrangement of sensors tailored to the specific activity of interest without the need for retraining. A novel autoregressive inferencing scheme ensures real-time motion reconstruction that closely aligns with measured sensor signals. The generative nature of DiffusionPoser ensures realistic behavior even for degrees-of-freedom not directly measured. Qualitative results can be found on our project website.

3D synthetic-to-real unsupervised domain adaptive segmentation is crucial to annotating new domains. Self-training is a competitive approach for this task but its performance is limited by different sensor sampling patterns (i.e. variations in point density) and incomplete training strategies. In this work we propose a density-guided translator (DGT) which translates point density between domains and integrates it into a two-stage self-training pipeline named DGT-ST. First in contrast to existing works that simultaneously conduct data generation and feature/output alignment within unstable adversarial training we employ the non-learnable DGT to bridge the domain gap at the input level. Second to provide a well-initialized model for self-training we propose a category-level adversarial network in stage one that utilizes the prototype to prevent negative transfer. Finally by leveraging the designs above a domain-mixed self-training method with source-aware consistency loss is proposed in stage two to narrow the domain gap further. Experiments on two synthetic-to-real segmentation tasks (SynLiDAR ? semanticKITTI and SynLiDAR ? semanticPOSS) demonstrate that DGT-ST outperforms state-of-the-art methods achieving 9.4% and 4.3% mIoU improvements respectively. Code is available at https://github.com/yuan-zm/DGT-ST.

The YOLO series has become the most popular framework for real-time object detection due to its reasonable trade-off between speed and accuracy. However we observe that the speed and accuracy of YOLOs are negatively affected by the NMS. Recently end-to-end Transformer-based detectors (DETRs) have provided an alternative to eliminating NMS. Nevertheless the high computational cost limits their practicality and hinders them from fully exploiting the advantage of excluding NMS. In this paper we propose the Real-Time DEtection TRansformer (RT-DETR) the first real-time end-to-end object detector to our best knowledge that addresses the above dilemma. We build RT-DETR in two steps drawing on the advanced DETR: first we focus on maintaining accuracy while improving speed followed by maintaining speed while improving accuracy. Specifically we design an efficient hybrid encoder to expeditiously process multi-scale features by decoupling intra-scale interaction and cross-scale fusion to improve speed. Then we propose the uncertainty-minimal query selection to provide high-quality initial queries to the decoder thereby improving accuracy. In addition RT-DETR supports flexible speed tuning by adjusting the number of decoder layers to adapt to various scenarios without retraining. Our RT-DETR-R50 / R101 achieves 53.1% / 54.3% AP on COCO and 108 / 74 FPS on T4 GPU outperforming previously advanced YOLOs in both speed and accuracy. We also develop scaled RT-DETRs that outperform the lighter YOLO detectors (S and M models). Furthermore RT-DETR-R50 outperforms DINO-R50 by 2.2% AP in accuracy and about 21 times in FPS. After pre-training with Objects365 RT-DETR-R50 / R101 achieves 55.3% / 56.2% AP. The project page: https://zhao-yian.github.io/RTDETR.

We introduce CyberDemo a novel approach to robotic imitation learning that leverages simulated human demonstrations for real-world tasks. By incorporating extensive data augmentation in a simulated environment CyberDemo outperforms traditional in-domain real-world demonstrations when transferred to the real world handling diverse physical and visual conditions. Regardless of its affordability and convenience in data collection CyberDemo outperforms baseline methods in terms of success rates across various tasks and exhibits generalizability with previously unseen objects. For example it can rotate novel tetra-valve and penta-valve despite human demonstrations only involving tri-valves. Our research demonstrates the significant potential of simulated human demonstrations for real world dexterous manipulation tasks. More details can be found at https://cyber-demo.github.io/

Quantized neural networks employ reduced precision representations for both weights and activations. This quantization process significantly reduces the memory requirements and computational complexity of the network. Binary Neural Networks (BNNs) are the extreme quantization case representing values with just one bit. Since the sign function is typically used to map real values to binary values smooth approximations are introduced to mimic the gradients during error backpropagation. Thus the mismatch between the forward and backward models corrupts the direction of the gradient causing training inconsistency problems and performance degradation. In contrast to current BNN approaches we propose to employ a binary periodic (BiPer) function during binarization. Specifically we use a square wave for the forward pass to obtain the binary values and employ the trigonometric sine function with the same period of the square wave as a differentiable surrogate during the backward pass. We demonstrate that this approach can control the quantization error by using the frequency of the periodic function and improves network performance. Extensive experiments validate the effectiveness of BiPer in benchmark datasets and network architectures with improvements of up to 1% and 0.69% with respect to state-of-the-art methods in the classification task over CIFAR-10 and ImageNet respectively. Our code is publicly available at https://github.com/edmav4/BiPer.

Implicit neural representation (INR) in combination with geometric rendering has recently been employed in real-time dense RGB-D SLAM. Despite active research endeavors being made there lacks a unified protocol for fair evaluation impeding the evolution of this area. In this work we establish to our knowledge the first open-source benchmark framework to evaluate the performance of a wide spectrum of commonly used INRs and rendering functions for mapping and localization. The goal of our benchmark is to 1) gain an intuition of how different INRs and rendering functions impact mapping and localization and 2) establish a unified evaluation protocol w.r.t. the design choices that may impact the mapping and localization. With the framework we conduct a large suite of experiments offering various insights in choosing the INRs and geometric rendering functions: for example the dense feature grid outperforms other INRs (e.g. tri-plane and hash grid) even when geometric and color features are jointly encoded for memory efficiency. To extend the findings into the practical scenario a hybrid encoding strategy is proposed to bring the best of the accuracy and completion from the grid-based and decomposition-based INRs. We further propose explicit hybrid encoding for high-fidelity dense grid mapping to comply with the RGB-D SLAM system that puts the premise on robustness and computation efficiency.

While real-world anime super-resolution (SR) has gained increasing attention in the SR community existing methods still adopt techniques from the photorealistic domain. In this paper we analyze the anime production workflow and rethink how to use characteristics of it for the sake of the real-world anime SR. First we argue that video networks and datasets are not necessary for anime SR due to the repetition use of hand-drawing frames. Instead we propose an anime image collection pipeline by choosing the least compressed and the most informative frames from the video sources. Based on this pipeline we introduce the Anime Production-oriented Image (API) dataset. In addition we identify two anime-specific challenges of distorted and faint hand-drawn lines and unwanted color artifacts. We address the first issue by introducing a prediction-oriented compression module in the image degradation model and a pseudo-ground truth preparation with enhanced hand-drawn lines. In addition we introduce the balanced twin perceptual loss combining both anime and photorealistic high-level features to mitigate unwanted color artifacts and increase visual clarity. We evaluate our method through extensive experiments on the public benchmark showing our method outperforms state-of-the-art anime dataset-trained approaches.

Generalization to new domains not seen during training is one of the long-standing challenges in deploying neural networks in real-world applications. Existing generalization techniques either necessitate external images for augmentation and/or aim at learning invariant representations by imposing various alignment constraints. Large-scale pretraining has recently shown promising generalization capabilities along with the potential of binding different modalities. For instance the advent of vision-language models like CLIP has opened the doorway for vision models to exploit the textual modality. In this paper we introduce a simple framework for generalizing semantic segmentation networks by employing language as the source of randomization. Our recipe comprises three key ingredients: (i) the preservation of the intrinsic CLIP robustness through minimal fine-tuning (ii) language-driven local style augmentation and (iii) randomization by locally mixing the source and augmented styles during training. Extensive experiments report state-of-the-art results on various generalization benchmarks.

This paper targets high-fidelity and real-time view synthesis of dynamic 3D scenes at 4K resolution. Recent methods on dynamic view synthesis have shown impressive rendering quality. However their speed is still limited when rendering high-resolution images. To overcome this problem we propose 4K4D a 4D point cloud representation that supports hardware rasterization and network pre-computation to enable unprecedented rendering speed with a high rendering quality. Our representation is built on a 4D feature grid so that the points are naturally regularized and can be robustly optimized. In addition we design a novel hybrid appearance model that significantly boosts the rendering quality while preserving efficiency. Moreover we develop a differentiable depth peeling algorithm to effectively learn the proposed model from RGB videos. Experiments show that our representation can be rendered at over 400 FPS on the DNA-Rendering dataset at 1080p resolution and 80 FPS on the ENeRF-Outdoor dataset at 4K resolution using an RTX 4090 GPU which is 30x faster than previous methods and achieves the state-of-the-art rendering quality. Our project page is available at https://zju3dv.github.io/4k4d.

Representing and rendering dynamic scenes has been an important but challenging task. Especially to accurately model complex motions high efficiency is usually hard to guarantee. To achieve real-time dynamic scene rendering while also enjoying high training and storage efficiency we propose 4D Gaussian Splatting (4D-GS) as a holistic representation for dynamic scenes rather than applying 3D-GS for each individual frame. In 4D-GS a novel explicit representation containing both 3D Gaussians and 4D neural voxels is proposed. A decomposed neural voxel encoding algorithm inspired by HexPlane is proposed to efficiently build Gaussian features from 4D neural voxels and then a lightweight MLP is applied to predict Gaussian deformations at novel timestamps. Our 4D-GS method achieves real-time rendering under high resolutions 82 FPS at an 800*800 resolution on an RTX 3090 GPU while maintaining comparable or better quality than previous state-of-the-art methods. More demos and code are available at https://guanjunwu.github.io/4dgs.