CVPR 2024 Papers

Semantic scene completion also known as semantic occupancy prediction can provide dense geometric and semantic information for autonomous vehicles which attracts the increasing attention of both academia and industry. Unfortunately existing methods usually formulate this task as a voxel-wise classification problem and treat each voxel equally in 3D space during training. As the hard voxels have not been paid enough attention the performance in some challenging regions is limited. The 3D dense space typically contains a large number of empty voxels which are easy to learn but require amounts of computation due to handling all the voxels uniformly for the existing models. Furthermore the voxels in the boundary region are more challenging to differentiate than those in the interior. In this paper we propose HASSC approach to train the semantic scene completion model with hardness-aware design. The global hardness from the network optimization process is defined for dynamical hard voxel selection. Then the local hardness with geometric anisotropy is adopted for voxel-wise refinement. Besides self-distillation strategy is introduced to make training process stable and consistent. Extensive experiments show that our HASSC scheme can effectively promote the accuracy of the baseline model without incurring the extra inference cost. Source code is available at: https://github.com/songw-zju/HASSC.

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.

The consistency training (CT)-based semi-supervised learning (SSL) bites state-of-the-art performance on SSL-based image classification. However the existing CT-based SSL methods do not highlight the non-Euclidean characteristics and class-wise varieties of embedding spaces in an SSL model thus they cannot fully utilize the effectiveness of CT. Thus we propose a metric tensor-based consistency regularization exploiting the class-variant geometrical structure of embeddings on the high-dimensional feature space. The proposed method not only minimizes the prediction discrepancy between different views of a given image but also estimates the intrinsic geometric curvature of embedding spaces by employing the global and local metric tensors. The global metric tensor is used to globally estimate the class-invariant embeddings from the whole data distribution while the local metric tensor is exploited to estimate the class-variant embeddings of each cluster. The two metric tensors are optimized by the consistency regularization based on the weak and strong augmentation strategy. The proposed method provides the highest classification accuracy on average compared to the existing state-of-the-art SSL methods on conventional datasets.

Normalizing flows have proven their efficacy for density estimation in Euclidean space but their application to rotational representations crucial in various domains such as robotics or human pose modeling remains underexplored. Probabilistic models of the human pose can benefit from approaches that rigorously consider the rotational nature of human joints. For this purpose we introduce HuProSO3 a normalizing flow model that operates on a high-dimensional product space of SO(3) manifolds modeling the joint distribution for human joints with three degrees of freedom. HuProSO3's advantage over state-of-the-art approaches is demonstrated through its superior modeling accuracy in three different applications and its capability to evaluate the exact likelihood. This work not only addresses the technical challenge of learning densities on SO(3) manifolds but it also has broader implications for domains where the probabilistic regression of correlated 3D rotations is of importance. Code will be available at https://github.com/odunkel/HuProSO.

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.

Even though Non-rigid Structure-from-Motion (NRSfM) has been extensively studied and great progress has been made there are still key challenges that hinder their broad real-world applications: 1) the inherent motion/rotation ambiguity requires either explicit camera motion recovery with extra constraint or complex Procrustean Alignment; 2) existing low-rank modeling of the global shape can over-penalize drastic deformations in the 3D shape sequence. This paper proposes to resolve the above issues from a spatial-temporal modeling perspective. First we propose a novel Temporally-smooth Procrustean Alignment module that estimates 3D deforming shapes and adjusts the camera motion by aligning the 3D shape sequence consecutively. Our new alignment module remedies the requirement of complex reference 3D shape during alignment which is more conductive to non-isotropic deformation modeling. Second we propose a spatial-weighted approach to enforce the low-rank constraint adaptively at different locations to accommodate drastic spatially-variant deformation reconstruction better. Our modeling outperform existing low-rank based methods and extensive experiments across different datasets validate the effectiveness of our method.

Text-to-image person re-identification (TIReID) is a compelling topic in the cross-modal community which aims to retrieve the target person based on a textual query. Although numerous TIReID methods have been proposed and achieved promising performance they implicitly assume the training image-text pairs are correctly aligned which is not always the case in real-world scenarios. In practice the image-text pairs inevitably exist under-correlated or even false-correlated a.k.a noisy correspondence (NC) due to the low quality of the images and annotation errors. To address this problem we propose a novel Robust Dual Embedding method (RDE) that can learn robust visual-semantic associations even with NC. Specifically RDE consists of two main components: 1) A Confident Consensus Division (CCD) module that leverages the dual-grained decisions of dual embedding modules to obtain a consensus set of clean training data which enables the model to learn correct and reliable visual-semantic associations. 2) A Triplet Alignment Loss (TAL) relaxes the conventional Triplet Ranking loss with the hardest negative samples to a log-exponential upper bound over all negative ones thus preventing the model collapse under NC and can also focus on hard-negative samples for promising performance. We conduct extensive experiments on three public benchmarks namely CUHK-PEDES ICFG-PEDES and RSTPReID to evaluate the performance and robustness of our RDE. Our method achieves state-of-the-art results both with and without synthetic noisy correspondences on all three datasets. Code is available at https://github.com/QinYang79/RDE.

Two-view homography estimation is a classic and fundamental problem in computer vision. While conceptually simple the problem quickly becomes challenging when multiple planes are visible in the image pair. Even with correct matches each individual plane (homography) might have a very low number of inliers when comparing to the set of all correspondences. In practice this requires a large number of RANSAC iterations to generate a good model hypothesis. The current state-of-the-art methods therefore seek to reduce the sample size from four point correspondences originally by including additional information such as keypoint orientation/angles or local affine information. In this work we continue in this direction and propose a novel one-point solver that leverages different approximate constraints derived from the same auxiliary information. In experiments we obtain state-of-the-art results with execution time speed-ups on large benchmark datasets and show that it is more beneficial for the solver to be sample efficient compared to generating more accurate homographies.

Layout-aware text-to-image generation is a task to generate multi-object images that reflect layout conditions in addition to text conditions. The current layout-aware text-to-image diffusion models still have several issues including mismatches between the text and layout conditions and quality degradation of generated images. This paper proposes a novel layout-aware text-to-image diffusion model called NoiseCollage to tackle these issues. During the denoising process NoiseCollage independently estimates noises for individual objects and then crops and merges them into a single noise. This operation helps avoid condition mismatches; in other words it can put the right objects in the right places. Qualitative and quantitative evaluations show that NoiseCollage outperforms several state-of-the-art models. These successful results indicate that the crop-and-merge operation of noises is a reasonable strategy to control image generation. We also show that NoiseCollage can be integrated with ControlNet to use edges sketches and pose skeletons as additional conditions. Experimental results show that this integration boosts the layout accuracy of ControlNet. The code is available at https://github.com/univ-esuty/noisecollage.

Generative models have been very popular in the recent years for their image generation capabilities. GAN-based models are highly regarded for their disentangled latent space which is a key feature contributing to their success in controlled image editing. On the other hand diffusion models have emerged as powerful tools for generating high-quality images. However the latent space of diffusion models is not as thoroughly explored or understood. Existing methods that aim to explore the latent space of diffusion models usually relies on text prompts to pinpoint specific semantics. However this approach may be restrictive in areas such as art fashion or specialized fields like medicine where suitable text prompts might not be available or easy to conceive thus limiting the scope of existing work. In this paper we propose an unsupervised method to discover latent semantics in text-to-image diffusion models without relying on text prompts. Our method takes a small set of unlabeled images from specific domains such as faces or cats and a pre-trained diffusion model and discovers diverse semantics in unsupervised fashion using a contrastive learning objective. Moreover the learned directions can be applied simultaneously either within the same domain (such as various types of facial edits) or across different domains (such as applying cat and face edits within the same image) without interfering with each other. Our extensive experiments show that our method achieves highly disentangled edits outperforming existing approaches in both diffusion-based and GAN-based latent space editing methods.

To reduce the reliance on large-scale datasets recent works in 3D segmentation resort to few-shot learning. Current 3D few-shot segmentation methods first pre-train models on 'seen' classes and then evaluate their generalization performance on 'unseen' classes. However the prior pre-training stage not only introduces excessive time overhead but also incurs a significant domain gap on 'unseen' classes. To tackle these issues we propose a Non-parametric Network for few-shot 3D Segmentation Seg-NN and its Parametric variant Seg-PN. Without training Seg-NN extracts dense representations by hand-crafted filters and achieves comparable performance to existing parameterized models. Due to the elimination of pre-training Seg-NN can alleviate the domain gap issue and save a substantial amount of time. Based on Seg-NN Seg-PN only requires training a lightweight QUEry-Support Transferring (QUEST) module which enhances the interaction between the support set and query set. Experiments suggest that Seg-PN outperforms previous state-of-the-art method by +4.19% and +7.71% mIoU on S3DIS and ScanNet datasets respectively while reducing training time by -90% indicating its effectiveness and efficiency. Code is available https://github.com/yangyangyang127/Seg-NN.

Inherent ambiguity in layout annotations poses significant challenges to developing accurate 360deg room layout estimation models. To address this issue we propose a novel Bi-Layout model capable of predicting two distinct layout types. One stops at ambiguous regions while the other extends to encompass all visible areas. Our model employs two global context embeddings where each embedding is designed to capture specific contextual information for each layout type. With our novel feature guidance module the image feature retrieves relevant context from these embeddings generating layout-aware features for precise bi-layout predictions. A unique property of our Bi-Layout model is its ability to inherently detect ambiguous regions by comparing the two predictions. To circumvent the need for manual correction of ambiguous annotations during testing we also introduce a new metric for disambiguating ground truth layouts. Our method demonstrates superior performance on benchmark datasets notably outperforming leading approaches. Specifically on the MatterportLayout dataset it improves 3DIoU from 81.70% to 82.57% across the full test set and notably from 54.80% to 59.97% in subsets with significant ambiguity.

Nighttime conditions pose a significant challenge to color constancy due to the diversity of lighting conditions and the presence of substantial low-light noise. Existing color constancy methods struggle with nighttime scenes frequently leading to imprecise light color estimations. To tackle nighttime color constancy we propose a novel unsupervised domain adaptation approach that utilizes labeled daytime data to facilitate learning on unlabeled nighttime images. To specifically address the unique lighting conditions of nighttime and ensure the robustness of pseudo labels we propose adaptive channel masking and light uncertainty. By selectively masking channels that are less sensitive to lighting conditions adaptive channel masking directs the model to progressively focus on features less affected by variations in light colors and noise. Additionally our model leverages light uncertainty to provide a pixel-wise uncertainty estimation regarding light color prediction which helps avoid learning from incorrect labels. Our model demonstrates a significant improvement in accuracy achieving 21.5% lower Mean Angular Error (MAE) compared to the state-of-the-art method on our nighttime dataset.

This paper presents Neural Visibility Field (NVF) a novel uncertainty quantification method for Neural Radiance Fields (NeRF) applied to active mapping. Our key insight is that regions not visible in the training views lead to inherently unreliable color predictions by NeRF at this region resulting in increased uncertainty in the synthesized views. To address this we propose to use Bayesian Networks to composite position-based field uncertainty into ray-based uncertainty in camera observations. Consequently NVF naturally assigns higher uncertainty to unobserved regions aiding robots to select the most informative next viewpoints. Extensive evaluations show that NVF excels not only in uncertainty quantification but also in scene reconstruction for active mapping outperforming existing methods. More details can be found at https://sites.google.com/view/nvf-cvpr24/.

The emerging conditional coding-based neural video codec (NVC) shows superiority over commonly-used residual coding-based codec and the latest NVC already claims to outperform the best traditional codec. However there still exist critical problems blocking the practicality of NVC. In this paper we propose a powerful conditional coding-based NVC that solves two critical problems via feature modulation. The first is how to support a wide quality range in a single model. Previous NVC with this capability only supports about 3.8 dB PSNR range on average. To tackle this limitation we modulate the latent feature of the current frame via the learnable quantization scaler. During the training we specially design the uniform quantization parameter sampling mechanism to improve the harmonization of encoding and quantization. This results in a better learning of the quantization scaler and helps our NVC support about 11.4 dB PSNR range. The second is how to make NVC still work under a long prediction chain. We expose that the previous SOTA NVC has an obvious quality degradation problem when using a large intra-period setting. To this end we propose modulating the temporal feature with a periodically refreshing mechanism to boost the quality. Notably under single intra-frame setting our codec can achieve 29.7% bitrate saving over previous SOTA NVC with 16% MACs reduction. Our codec serves as a notable landmark in the journey of NVC evolution. The codes are at https://github.com/microsoft/DCVC.

Among the numerous efforts towards digitally recovering the physical world Neural Radiance Fields (NeRFs) have proved effective in most cases. However underwater scene introduces unique challenges due to the absorbing water medium the local change in lighting and the dynamic contents in the scene. We aim at developing a neural underwater scene representation for these challenges modeling the complex process of attenuation unstable in-scattering and moving objects during light transport. The proposed method can reconstruct the scenes from both established datasets and in-the-wild videos with outstanding fidelity.

It is time-consuming to render high-resolution images in applications such as video games and virtual reality and thus super-resolution technologies become increasingly popular for real-time rendering. However it is challenging to preserve sharp texture details keep the temporal stability and avoid the ghosting artifacts in real-time super-resolution rendering. To address this issue we introduce radiance demodulation to separate the rendered image or radiance into a lighting component and a material component considering the fact that the light component is smoother than the rendered image so that the high-resolution material component with detailed textures can be easily obtained. We perform the super-resolution on the lighting component only and re-modulate it with the high-resolution material component to obtain the final super-resolution image with more texture details. A reliable warping module is proposed by explicitly marking the occluded regions to avoid the ghosting artifacts. To further enhance the temporal stability we design a frame-recurrent neural network and a temporal loss to aggregate the previous and current frames which can better capture the spatial-temporal consistency among reconstructed frames. As a result our method is able to produce temporally stable results in real-time rendering with high-quality details even in the challenging 4 x4 super-resolution scenarios. Code is available at: \href https://github.com/Riga2/NSRD https://github.com/Riga2/NSRD .

Each photo in an image burst can be considered a sample of a complex 3D scene: the product of parallax diffuse and specular materials scene motion and illuminant variation. While decomposing all of these effects from a stack of misaligned images is a highly ill-conditioned task the conventional align-and-merge burst pipeline takes the other extreme: blending them into a single image. In this work we propose a versatile intermediate representation: a two-layer alpha-composited image plus flow model constructed with neural spline fields -- networks trained to map input coordinates to spline control points. Our method is able to during test-time optimization jointly fuse a burst image capture into one high-resolution reconstruction and decompose it into transmission and obstruction layers. Then by discarding the obstruction layer we can perform a range of tasks including seeing through occlusions reflection suppression and shadow removal. Tested on complex in-the-wild captures we find that with no post-processing steps or learned priors our generalizable model is able to outperform existing dedicated single-image and multi-view obstruction removal approaches.

Sign Languages (SL) serve as the primary mode of communication for the Deaf and Hard of Hearing communities. Deep learning methods for SL recognition and translation have achieved promising results. However Sign Language Production (SLP) poses a challenge as the generated motions must be realistic and have precise semantic meaning. Most SLP methods rely on 2D data which hinders their realism. In this work a diffusion-based SLP model is trained on a curated large-scale dataset of 4D signing avatars and their corresponding text transcripts. The proposed method can generate dynamic sequences of 3D avatars from an unconstrained domain of discourse using a diffusion process formed on a novel and anatomically informed graph neural network defined on the SMPL-X body skeleton. Through quantitative and qualitative experiments we show that the proposed method considerably outperforms previous methods of SLP. This work makes an important step towards realistic neural sign avatars bridging the communication gap between Deaf and hearing communities.