NeurIPS 2024 Papers

Recent advances in implicit neural representation (INR)-based video coding have demonstrated its potential to compete with both conventional and other learning- based approaches. With INR methods, a neural network is trained to overfit a video sequence, with its parameters compressed to obtain a compact representation of the video content. However, although promising results have been achieved, the best INR-based methods are still out-performed by the latest standard codecs, such as VVC VTM, partially due to the simple model compression techniques employed. In this paper, rather than focusing on representation architectures, which is a common focus in many existing works, we propose a novel INR-based video compression framework, Neural Video Representation Compression (NVRC), targeting compression of the representation. Based on its novel quantization and entropy coding approaches, NVRC is the first framework capable of optimizing an INR-based video representation in a fully end-to-end manner for the rate-distortion trade-off. To further minimize the additional bitrate overhead introduced by the entropy models, NVRC also compresses all the network, quantization and entropy model parameters hierarchically. Our experiments show that NVRC outperforms many conventional and learning-based benchmark codecs, with a 23% average coding gain over VVC VTM (Random Access) on the UVG dataset, measured in PSNR. As far as we are aware, this is the first time an INR-based video codec achieving such performance.

Visual object tracking aims to locate a targeted object in a video sequence based on an initial bounding box. Recently, Vision-Language~(VL) trackers have proposed to utilize additional natural language descriptions to enhance versatility in various applications. However, VL trackers are still inferior to State-of-The-Art (SoTA) visual trackers in terms of tracking performance. We found that this inferiority primarily results from their heavy reliance on manual textual annotations, which include the frequent provision of ambiguous language descriptions. In this paper, we propose ChatTracker to leverage the wealth of world knowledge in the Multimodal Large Language Model (MLLM) to generate high-quality language descriptions and enhance tracking performance. To this end, we propose a novel reflection-based prompt optimization module to iteratively refine the ambiguous and inaccurate descriptions of the target with tracking feedback. To further utilize semantic information produced by MLLM, a simple yet effective VL tracking framework is proposed and can be easily integrated as a plug-and-play module to boost the performance of both VL and visual trackers. Experimental results show that our proposed ChatTracker achieves a performance comparable to existing methods.

Quanta image sensors, such as SPAD arrays, are an emerging sensor technology, producing 1-bit arrays representing photon detection events over exposures as short as a few nanoseconds. In practice, raw data are post-processed using heavy spatiotemporal binning to create more useful and interpretable images at the cost of degrading spatiotemporal resolution. In this work, we propose bit2bit, a new method for reconstructing high-quality image stacks at the original spatiotemporal resolution from sparse binary quanta image data. Inspired by recent work on Poisson denoising, we developed an algorithm that creates a dense image sequence from sparse binary photon data by predicting the photon arrival location probability distribution. However, due to the binary nature of the data, we show that the assumption of a Poisson distribution is inadequate. Instead, we model the process with a Bernoulli lattice process from the truncated Poisson. This leads to the proposal of a novel self-supervised solution based on a masked loss function. We evaluate our method using both simulated and real data. On simulated data from a conventional video, we achieve 34.35 mean PSNR with extremely photon-sparse binary input (<0.06 photons per pixel per frame). We also present a novel dataset containing a wide range of real SPAD high-speed videos under various challenging imaging conditions. The scenes cover strong/weak ambient light, strong motion, ultra-fast events, etc., which will be made available to the community, on which we demonstrate the promise of our approach. Both reconstruction quality and throughput substantially surpass the state-of-the-art methods (e.g., Quanta Burst Photography (QBP)). Our approach significantly enhances the visualization and usability of the data, enabling the application of existing analysis techniques.

High dynamic range (HDR) novel view synthesis (NVS) aims to create photorealistic images from novel viewpoints using HDR imaging techniques. The rendered HDR images capture a wider range of brightness levels containing more details of the scene than normal low dynamic range (LDR) images. Existing HDR NVS methods are mainly based on NeRF. They suffer from long training time and slow inference speed. In this paper, we propose a new framework, High Dynamic Range Gaussian Splatting (HDR-GS), which can efficiently render novel HDR views and reconstruct LDR images with a user input exposure time. Specifically, we design a Dual Dynamic Range (DDR) Gaussian point cloud model that uses spherical harmonics to fit HDR color and employs an MLP-based tone-mapper to render LDR color. The HDR and LDR colors are then fed into two Parallel Differentiable Rasterization (PDR) processes to reconstruct HDR and LDR views. To establish the data foundation for the research of 3D Gaussian splatting-based methods in HDR NVS, we recalibrate the camera parameters and compute the initial positions for Gaussian point clouds. Comprehensive experiments show that HDR-GS surpasses the state-of-the-art NeRF-based method by 3.84 and 1.91 dB on LDR and HDR NVS while enjoying 1000$\times$ inference speed and only costing 6.3\% training time. Code and data are released at https://github.com/caiyuanhao1998/HDR-GS

It is important to estimate an accurate signed distance function (SDF) from a point cloud in many computer vision applications. The latest methods learn neural SDFs using either a data-driven based or an overfitting-based strategy. However, these two kinds of methods are with either poor generalization or slow convergence, which limits their capability under challenging scenarios like highly noisy point clouds. To resolve this issue, we propose a method to prompt pros of both data-driven based and overfitting-based methods for better generalization, faster inference, and higher accuracy in learning neural SDFs. We introduce a novel statistical reasoning algorithm in local regions which is able to finetune data-driven based priors without signed distance supervision, clean point cloud, or point normals. This helps our method start with a good initialization, and converge to a minimum in a much faster way. Our numerical and visual comparisons with the stat-of-the-art methods show our superiority over these methods in surface reconstruction and point cloud denoising on widely used shape and scene benchmarks. The code is available at https://github.com/chenchao15/LocalN2NM.

Developing meaningful and efficient representations that separate the fundamental structure of the data generation mechanism is crucial in representation learning. However, Disentangled Representation Learning has not fully shown its potential on real images, because of correlated generative factors, their resolution and limited access to ground truth labels. Specifically on the latter, we investigate the possibility of leveraging synthetic data to learn general-purpose disentangled representations applicable to real data, discussing the effect of fine-tuning and what properties of disentanglement are preserved after the transfer. We provide an extensive empirical study to address these issues. In addition, we propose a new interpretable intervention-based metric, to measure the quality of factors encoding in the representation. Our results indicate that some level of disentanglement, transferring a representation from synthetic to real data, is possible and effective.

Image Restoration (IR) methods based on a pre-trained diffusion model have demonstrated state-of-the-art performance. However, they have two fundamental limitations: 1) they often assume that the degradation operator is completely known and 2) they alter the diffusion sampling process, which may result in restored images that do not lie onto the data manifold. To address these issues, we propose Blind Image Restoration via fast Diffusion inversion (BIRD) a blind IR method that jointly optimizes for the degradation model parameters and the restored image. To ensure that the restored images lie onto the data manifold, we propose a novel sampling technique on a pre-trained diffusion model. A key idea in our method is not to modify the reverse sampling, i.e., not to alter all the intermediate latents, once an initial noise is sampled. This is ultimately equivalent to casting the IR task as an optimization problem in the space of the input noise. Moreover, to mitigate the computational cost associated with inverting a fully unrolled diffusion model, we leverage the inherent capability of these models to skip ahead in the forward diffusion process using large time steps. We experimentally validate BIRD on several image restoration tasks and show that it achieves state of the art performance.

Detecting text generated by Large Language Models (LLMs) is a pressing need in order to identify and prevent misuse of these powerful models in a wide range of applications, which have highly undesirable consequences such as misinformation and academic dishonesty. Given a piece of subject text, many existing detection methods work by measuring the difficulty of LLM predicting the next token in the text from their prefix. In this paper, we make a critical observation that how well the current token’s output logits memorizes the closely preceding input tokens also provides strong evidence. Therefore, we propose a novel bi-directional calculation method that measures the cross-entropy losses between an output logits and the ground-truth token (forward) and between the output logits and the immediately preceding input token (backward). A classifier is trained to make the final prediction based on the statistics of these losses. We evaluate our system, named BISCOPE, on texts generated by five latest commercial LLMs across five heterogeneous datasets, including both natural language and code. BISCOPE demonstrates superior detection accuracy and robustness compared to six existing baseline methods, exceeding the state-of-the-art non-commercial methods’ detection accuracy by over 0.30 F1 score, achieving over 0.95 detection F1 score on average. It also outperforms the best commercial tool GPTZero that is based on a commercial LLM trained with an enormous volume of data. Code is available at https://github.com/MarkGHX/BiScope.

Stereo matching algorithms that leverage end-to-end convolutional neural networks have recently demonstrated notable advancements in performance. However, a common issue is their susceptibility to domain shifts, hindering their ability in generalizing to diverse, unseen realistic domains. We argue that existing stereo matching networks overlook the importance of extracting semantically and structurally meaningful features. To address this gap, we propose an effective hierarchical object-aware dual-level contrastive learning (HODC) framework for domain generalized stereo matching. Our framework guides the model in extracting features that support semantically and structurally driven matching by segmenting objects at different scales and enhances correspondence between intra- and inter-scale regions from the left feature map to the right using dual-level contrastive loss. HODC can be integrated with existing stereo matching models in the training stage, requiring no modifications to the architecture. Remarkably, using only synthetic datasets for training, HODC achieves state-of-the-art generalization performance with various existing stereo matching network architectures, across multiple realistic datasets.

The visual medium (images and videos) naturally contains a large amount of information redundancy, thereby providing a great opportunity for leveraging efficiency in processing. While Vision Transformer (ViT) based models scale effectively to large data regimes, they fail to capitalize on this inherent redundancy, leading to higher computational costs. Mixture of Experts (MoE) networks demonstrate scalability while maintaining same inference-time costs, but they come with a larger parameter footprint. We present Mixture of Nested Experts (MoNE), which utilizes a nested structure for experts, wherein individual experts fall on an increasing compute-accuracy curve. Given a compute budget, MoNE learns to dynamically choose tokens in a priority order, and thus redundant tokens are processed through cheaper nested experts. Using this framework, we achieve equivalent performance as the baseline models, while reducing inference time compute by over two-fold. We validate our approach on standard image and video datasets - ImageNet-21K, Kinetics400, and Something-Something-v2. We further highlight MoNE's adaptability by showcasing its ability to maintain strong performance across different inference-time compute budgets on videos, using only a single trained model.

The recent advances in 3D Gaussian Splatting (3DGS) show promising results on the novel view synthesis (NVS) task. With its superior rendering performance and high-fidelity rendering quality, 3DGS is excelling at its previous NeRF counterparts. The most recent 3DGS method focuses either on improving the instability of rendering efficiency or reducing the model size. On the other hand, the training efficiency of 3DGS on large-scale scenes has not gained much attention. In this work, we propose DoGaussian, a method that trains 3DGS distributedly. Our method first decomposes a scene into $K$ blocks and then introduces the Alternating Direction Method of Multipliers (ADMM) into the training procedure of 3DGS. During training, our DoGaussian maintains one global 3DGS model on the master node and $K$ local 3DGS models on the slave nodes. The $K$ local 3DGS models are dropped after training and we only query the global 3DGS model during inference. The training time is reduced by scene decomposition, and the training convergence and stability are guaranteed through the consensus on the shared 3D Gaussians. Our method accelerates the training of 3DGS by $6+$ times when evaluated on large-scale scenes while concurrently achieving state-of-the-art rendering quality. Our code is publicly available at [https://github.com/AIBluefisher/DOGS](https://github.com/AIBluefisher/DOGS).

The pre-trained point cloud model based on Masked Point Modeling (MPM) has exhibited substantial improvements across various tasks. However, these models heavily rely on the Transformer, leading to quadratic complexity and limited decoder, hindering their practice application. To address this limitation, we first conduct a comprehensive analysis of existing Transformer-based MPM, emphasizing the idea that redundancy reduction is crucial for point cloud analysis. To this end, we propose a Locally constrained Compact point cloud Model (LCM) consisting of a locally constrained compact encoder and a locally constrained Mamba-based decoder. Our encoder replaces self-attention with our local aggregation layers to achieve an elegant balance between performance and efficiency. Considering the varying information density between masked and unmasked patches in the decoder inputs of MPM, we introduce a locally constrained Mamba-based decoder. This decoder ensures linear complexity while maximizing the perception of point cloud geometry information from unmasked patches with higher information density. Extensive experimental results show that our compact model significantly surpasses existing Transformer-based models in both performance and efficiency, especially our LCM-based Point-MAE model, compared to the Transformer-based model, achieved an improvement of 1.84%, 0.67%, and 0.60% in performance on the three variants of ScanObjectNN while reducing parameters by 88% and computation by 73%. The code is available at https://github.com/zyh16143998882/LCM.

The key challenge of multi-view indoor 3D object detection is to infer accurate geometry information from images for precise 3D detection. Previous method relies on NeRF for geometry reasoning. However, the geometry extracted from NeRF is generally inaccurate, which leads to sub-optimal detection performance. In this paper, we propose MVSDet which utilizes plane sweep for geometry-aware 3D object detection. To circumvent the requirement for a large number of depth planes for accurate depth prediction, we design a probabilistic sampling and soft weighting mechanism to decide the placement of pixel features on the 3D volume. We select multiple locations that score top in the probability volume for each pixel and use their probability score to indicate the confidence. We further apply recent pixel-aligned Gaussian Splatting to regularize depth prediction and improve detection performance with little computation overhead. Extensive experiments on ScanNet and ARKitScenes datasets are conducted to show the superiority of our model. Our code is available at https://github.com/Pixie8888/MVSDet.

We study the problem of precisely swapping objects in videos, with a focus on those interacted with by hands, given one user-provided reference object image. Despite the great advancements that diffusion models have made in video editing recently, these models often fall short in handling the intricacies of hand-object interactions (HOI), failing to produce realistic edits---especially when object swapping results in object shape or functionality changes. To bridge this gap, we present HOI-Swap, a novel diffusion-based video editing framework trained in a self-supervised manner. Designed in two stages, the first stage focuses on object swapping in a single frame with HOI awareness; the model learns to adjust the interaction patterns, such as the hand grasp, based on changes in the object's properties. The second stage extends the single-frame edit across the entire sequence; we achieve controllable motion alignment with the original video by: (1) warping a new sequence from the stage-I edited frame based on sampled motion points and (2) conditioning video generation on the warped sequence. Comprehensive qualitative and quantitative evaluations demonstrate that HOI-Swap significantly outperforms existing methods, delivering high-quality video edits with realistic HOIs.

Heatmap regression has dominated human pose estimation due to its superior performance and strong generalization. To meet the requirements of traditional explicit neural networks for output form, existing heatmap-based methods discretize the originally continuous heatmap representation into 2D pixel arrays, which leads to performance degradation due to the introduction of quantization errors. This problem is significantly exacerbated as the size of the input image decreases, which makes heatmap-based methods not much better than coordinate regression on low-resolution images. In this paper, we propose a novel neural representation for human pose estimation called NerPE to achieve continuous heatmap regression. Given any position within the image range, NerPE regresses the corresponding confidence scores for body joints according to the surrounding image features, which guarantees continuity in space and confidence during training. Thanks to the decoupling from spatial resolution, NerPE can output the predicted heatmaps at arbitrary resolution during inference without retraining, which easily achieves sub-pixel localization precision. To reduce the computational cost, we design progressive coordinate decoding to cooperate with continuous heatmap regression, in which localization no longer requires the complete generation of high-resolution heatmaps. The code is available at https://github.com/hushengxiang/NerPE.

Recently, federated multi-view clustering (FedMVC) has emerged to explore cluster structures in multi-view data distributed on multiple clients. Many existing approaches tend to assume that clients are isomorphic and all of them belong to either single-view clients or multi-view clients. While these methods have succeeded, they may encounter challenges in practical FedMVC scenarios involving heterogeneous hybrid views, where a mixture of single-view and multi-view clients exhibit varying degrees of heterogeneity. In this paper, we propose a novel FedMVC framework, which concurrently addresses two challenges associated with heterogeneous hybrid views, i.e., client gap and view gap. To address the client gap, we design a local-synergistic contrastive learning approach that helps single-view clients and multi-view clients achieve consistency for mitigating heterogeneity among all clients. To address the view gap, we develop a global-specific weighting aggregation method, which encourages global models to learn complementary features from hybrid views. The interplay between local-synergistic contrastive learning and global-specific weighting aggregation mutually enhances the exploration of the data cluster structures distributed on multiple clients. Theoretical analysis and extensive experiments demonstrate that our method can handle the heterogeneous hybrid views in FedMVC and outperforms state-of-the-art methods.

Volume rendering in neural radiance fields is inherently time-consuming due to the large number of MLP calls on the points sampled per ray. Previous works would address this issue by introducing new neural networks or data structures. In this work, we propose GL-NeRF, a new perspective of computing volume rendering with the Gauss-Laguerre quadrature. GL-NeRF significantly reduces the number of MLP calls needed for volume rendering, introducing no additional data structures or neural networks. The simple formulation makes adopting GL-NeRF in any NeRF model possible. In the paper, we first justify the use of the Gauss-Laguerre quadrature and then demonstrate this plug-and-play attribute by implementing it in two different NeRF models. We show that with a minimal drop in performance, GL-NeRF can significantly reduce the number of MLP calls, showing the potential to speed up any NeRF model. Code can be found in project page https://silongyong.github.io/GL-NeRF_project_page/.

Cross-domain few-shot segmentation (CD-FSS) is proposed to first pre-train the model on a large-scale source-domain dataset, and then transfer the model to data-scarce target-domain datasets for pixel-level segmentation. The significant domain gap between the source and target datasets leads to a sharp decline in the performance of existing few-shot segmentation (FSS) methods in cross-domain scenarios. In this work, we discover an intriguing phenomenon: simply filtering different frequency components for target domains can lead to a significant performance improvement, sometimes even as high as 14% mIoU. Then, we delve into this phenomenon for an interpretation, and find such improvements stem from the reduced inter-channel correlation in feature maps, which benefits CD-FSS with enhanced robustness against domain gaps and larger activated regions for segmentation. Based on this, we propose a lightweight frequency masker, which further reduces channel correlations by an Amplitude-Phase Masker (APM) module and an Adaptive Channel Phase Attention (ACPA) module. Notably, APM introduces only 0.01% additional parameters but improves the average performance by over 10%, and ACPA imports only 2.5% parameters but further improves the performance by over 1.5%, which significantly surpasses the state-of-the-art CD-FSS methods.

Monocular depth estimation (MDE) is fundamental for deriving 3D scene structures from 2D images. While state-of-the-art monocular relative depth estimation (MRDE) excels in estimating relative depths for in-the-wild images, current monocular metric depth estimation (MMDE) approaches still face challenges in handling unseen scenes. Since MMDE can be viewed as the composition of MRDE and metric scale recovery, we attribute this difficulty to scene dependency, where MMDE models rely on scenes observed during supervised training for predicting scene scales during inference. To address this issue, we propose to use humans as landmarks for distilling scene-independent metric scale priors from generative painting models. Our approach, Metric from Human (MfH), bridges from generalizable MRDE to zero-shot MMDE in a generate-and-estimate manner. Specifically, MfH generates humans on the input image with generative painting and estimates human dimensions with an off-the-shelf human mesh recovery (HMR) model. Based on MRDE predictions, it propagates the metric information from painted humans to the contexts, resulting in metric depth estimations for the original input. Through this annotation-free test-time adaptation, MfH achieves superior zero-shot performance in MMDE, demonstrating its strong generalization ability.