NeurIPS 2024 Papers

Image feature extractors are rendered substantially more useful if different views of the same 3D location yield similar features while still being distinct from other locations. A feature extractor that achieves this goal even under significant viewpoint changes must recognise not just semantic categories in a scene, but also understand how different objects relate to each other in three dimensions. Existing work addresses this task by posing it as a patch retrieval problem, training the extracted features to facilitate retrieval of all image patches that project from the same 3D location. However, this approach uses a loss formulation that requires substantial memory and computation resources, limiting its applicability for large-scale training. We present a method for memory-efficient learning of location-consistent features that reformulates and approximates the smooth average precision objective. This novel loss function enables improvements in memory efficiency by three orders of magnitude, mitigating a key bottleneck of previous methods and allowing much larger models to be trained with the same computational resources. We showcase the improved location consistency of our trained feature extractor directly on a multi-view consistency task, as well as the downstream task of scene-stable panoptic segmentation, significantly outperforming previous state-of-the-art.

The metric backbone of a weighted graph is the union of all-pairs shortest paths. It is obtained by removing all edges $(u,v)$ that are not the shortest path between $u$ and $v$. In networks with well-separated communities, the metric backbone tends to preserve many inter-community edges, because these edges serve as bridges connecting two communities, but tends to delete many intra-community edges because the communities are dense. This suggests that the metric backbone would dilute or destroy the community structure of the network. However, this is not borne out by prior empirical work, which instead showed that the metric backbone of real networks preserves the community structure of the original network well. In this work, we analyze the metric backbone of a broad class of weighted random graphs with communities, and we formally prove the robustness of the community structure with respect to the deletion of all the edges that are not in the metric backbone. An empirical comparison of several graph sparsification techniques confirms our theoretical finding and shows that the metric backbone is an efficient sparsifier in the presence of communities.

Blended-target domain adaptation (BTDA), which implicitly mixes multiple sub-target domains into a fine domain, has attracted more attention in recent years. Most previously developed BTDA approaches focus on utilizing a single source domain, which makes it difficult to obtain sufficient feature information for learning domain-invariant representations. Furthermore, different feature distributions derived from different domains may increase the uncertainty of models. To overcome these issues, we propose a style adaptation and uncertainty estimation (SAUE) approach for multi-source blended-target domain adaptation (MBDA). Specifically, we exploit the extra knowledge acquired from the blended-target domain, where a similarity factor is adopted to select more useful target style information for augmenting the source features. \!Then, to mitigate the negative impact of the domain-specific attributes, we devise a function to estimate and mitigate uncertainty in category prediction. Finally, we construct a simple and lightweight adversarial learning strategy for MBDA, effectively aligning multi-source and blended-target domains without the requirements of domain labels of the target domains. Extensive experiments conducted on several challenging DA benchmarks, including the ImageCLEF-DA, Office-Home, VisDA 2017, and DomainNet datasets, demonstrate the superiority of our method over the state-of-the-art (SOTA) approaches.

The field of face recognition (FR) has undergone significant advancements with the rise of deep learning. Recently, the success of unsupervised learning and graph neural networks has demonstrated the effectiveness of data structure information. Considering that the FR task can leverage large-scale training data, which intrinsically contains significant structure information, we aim to investigate how to encode such critical structure information into the latent space. As revealed from our observations, directly aligning the structure information between the input and latent spaces inevitably suffers from an overfitting problem, leading to a structure collapse phenomenon in the latent space. To address this problem, we propose TopoFR, a novel FR model that leverages a topological structure alignment strategy called PTSA and a hard sample mining strategy named SDE. Concretely, PTSA uses persistent homology to align the topological structures of the input and latent spaces, effectively preserving the structure information and improving the generalization performance of FR model. To mitigate the impact of hard samples on the latent space structure, SDE accurately identifies hard samples by automatically computing structure damage score (SDS) for each sample, and directs the model to prioritize optimizing these samples. Experimental results on popular face benchmarks demonstrate the superiority of our TopoFR over the state-of-the-art methods. Code and models are available at: https://github.com/modelscope/facechain/tree/main/face_module/TopoFR.

Large vision language models (LVLMs) often suffer from object hallucination, producing objects not present in the given images. While current benchmarks for object hallucination primarily concentrate on the presence of a single object class rather than individual entities, this work systematically investigates multi-object hallucination, examining how models misperceive (e.g., invent nonexistent objects or become distracted) when tasked with focusing on multiple objects simultaneously. We introduce Recognition-based Object Probing Evaluation (ROPE), an automated evaluation protocol that considers the distribution of object classes within a single image during testing and uses visual referring prompts to eliminate ambiguity. With comprehensive empirical studies and analysis of potential factors leading to multi-object hallucination, we found that (1) LVLMs suffer more hallucinations when focusing on multiple objects compared to a single object. (2) The tested object class distribution affects hallucination behaviors, indicating that LVLMs may follow shortcuts and spurious correlations. (3) Hallucinatory behaviors are influenced by data-specific factors, salience and frequency, and model intrinsic behaviors. We hope to enable LVLMs to recognize and reason about multiple objects that often occur in realistic visual scenes, provide insights, and quantify our progress towards mitigating the issues.

This paper introduces Chain-of-Sight, a vision-language bridge module that accelerates the pre-training of Multimodal Large Language Models (MLLMs). Our approach employs a sequence of visual resamplers that capture visual details at various spacial scales. This architecture not only leverages global and local visual contexts effectively, but also facilitates the flexible extension of visual tokens through a compound token scaling strategy, allowing up to a 16x increase in the token count post pre-training. Consequently, Chain-of-Sight requires significantly fewer visual tokens in the pre-training phase compared to the fine-tuning phase. This intentional reduction of visual tokens during pre-training notably accelerates the pre-training process, cutting down the wall-clock training time by $\sim$73\%. Empirical results on a series of vision-language benchmarks reveal that the pre-train acceleration through Chain-of-Sight is achieved without sacrificing performance, matching or surpassing the standard pipeline of utilizing all visual tokens throughout the entire training process. Further scaling up the number of visual tokens for pre-training leads to stronger performances, competitive to existing approaches in a series of benchmarks.

Neural radiance fields (NeRFs) have gained popularity with multiple works showing promising results across various applications. However, to the best of our knowledge, existing works do not explicitly model the distribution of training camera poses, or consequently the triangulation quality, a key factor affecting reconstruction quality dating back to classical vision literature. We close this gap with ProvNeRF, an approach that models the provenance for each point -- i.e., the locations where it is likely visible -- of NeRFs as a stochastic field. We achieve this by extending implicit maximum likelihood estimation (IMLE) to functional space with an optimizable objective. We show that modeling per-point provenance during the NeRF optimization enriches the model with information on triangulation leading to improvements in novel view synthesis and uncertainty estimation under the challenging sparse, unconstrained view setting against competitive baselines. The code will be available at https://github.com/georgeNakayama/ProvNeRF.

The ability to detect aerial objects with limited annotation is pivotal to the development of real-world aerial intelligence systems. In this work, we focus on a demanding but practical sparsely annotated object detection (SAOD) in aerial images, which encompasses a wider variety of aerial scenes with the same number of annotated objects. Although most existing SAOD methods rely on fixed thresholding to filter pseudo-labels for enhancing detector performance, adapting to aerial objects proves challenging due to the imbalanced probabilities/confidences associated with predicted aerial objects. To address this problem, we propose a novel Progressive Exploration-Conformal Learning (PECL) framework to address the SAOD task, which can adaptively perform the selection of high-quality pseudo-labels in aerial images. Specifically, the pseudo-label exploration can be formulated as a decision-making paradigm by adopting a conformal pseudo-label explorer and a multi-clue selection evaluator. The conformal pseudo-label explorer learns an adaptive policy by maximizing the cumulative reward, which can decide how to select these high-quality candidates by leveraging their essential characteristics and inter-instance contextual information. The multi-clue selection evaluator is designed to evaluate the explorer-guided pseudo-label selections by providing an instructive feedback for policy optimization. Finally, the explored pseudo-labels can be adopted to guide the optimization of aerial object detector in a closed-looping progressive fashion. Comprehensive evaluations on two public datasets demonstrate the superiority of our PECL when compared with other state-of-the-art methods in the sparsely annotated aerial object detection task.

In recent advancements in unsupervised visual representation learning, the Joint-Embedding Predictive Architecture (JEPA) has emerged as a significant method for extracting visual features from unlabeled imagery through an innovative masking strategy. Despite its success, two primary limitations have been identified: the inefficacy of Exponential Moving Average (EMA) from I-JEPA in preventing entire collapse and the inadequacy of I-JEPA prediction in accurately learning the mean of patch representations. Addressing these challenges, this study introduces a novel framework, namely C-JEPA (Contrastive-JEPA), which integrates the Image-based Joint-Embedding Predictive Architecture with the Variance-Invariance-Covariance Regularization (VICReg) strategy. This integration is designed to effectively learn the variance/covariance for preventing entire collapse and ensuring invariance in the mean of augmented views, thereby overcoming the identified limitations. Through empirical and theoretical evaluations, our work demonstrates that C-JEPA significantly enhances the stability and quality of visual representation learning. When pre-trained on the ImageNet-1K dataset, C-JEPA exhibits rapid and improved convergence in both linear probing and fine-tuning performance metrics.

This paper scales object-level reconstruction to complex scenes, advancing interactive scene reconstruction. We introduce two datasets, OmniSim and InterReal, featuring 28 scenes with multiple interactive objects. To tackle the challenge of inaccurate interactive motion recovery in complex scenes, we propose LiveScene, a scene-level language-embedded interactive radiance field that efficiently reconstructs and controls multiple objects. By decomposing the interactive scene into local deformable fields, LiveScene enables separate reconstruction of individual object motions, reducing memory consumption. Additionally, our interaction-aware language embedding localizes individual interactive objects, allowing for arbitrary control using natural language. Our approach demonstrates significant superiority in novel view synthesis, interactive scene control, and language grounding performance through extensive experiments. Project page: https://livescenes.github.io.

Despite alleviating the dependence on dense annotations inherent to fully supervised methods, weakly supervised point cloud semantic segmentation suffers from inadequate supervision signals. In response to this challenge, we introduce a novel perspective that imparts auxiliary constraints by regulating the feature space under weak supervision. Our initial investigation identifies which distributions accurately characterize the feature space, subsequently leveraging this priori to guide the alignment of the weakly supervised embeddings. Specifically, we analyze the superiority of the mixture of von Mises-Fisher distributions (moVMF) among several common distribution candidates. Accordingly, we develop a Distribution Guidance Network (DGNet), which comprises a weakly supervised learning branch and a distribution alignment branch. Leveraging reliable clustering initialization derived from the weakly supervised learning branch, the distribution alignment branch alternately updates the parameters of the moVMF and the network, ensuring alignment with the moVMF-defined latent space. Extensive experiments validate the rationality and effectiveness of our distribution choice and network design. Consequently, DGNet achieves state-of-the-art performance under multiple datasets and various weakly supervised settings.

Controllable synthetic data generation can substantially lower the annotation cost of training data. Prior works use diffusion models to generate driving images conditioned on the 3D object layout. However, those models are trained on small-scale datasets like nuScenes, which lack appearance and layout diversity. Moreover, overfitting often happens, where the trained models can only generate images based on the layout data from the validation set of the same dataset. In this work, we introduce a simulator-conditioned scene generation framework called SimGen that can learn to generate diverse driving scenes by mixing data from the simulator and the real world. It uses a novel cascade diffusion pipeline to address challenging sim-to-real gaps and multi-condition conflicts. A driving video dataset DIVA is collected to enhance the generative diversity of SimGen, which contains over 147.5 hours of real-world driving videos from 73 locations worldwide and simulated driving data from the MetaDrive simulator. SimGen achieves superior generation quality and diversity while preserving controllability based on the text prompt and the layout pulled from a simulator. We further demonstrate the improvements brought by SimGen for synthetic data augmentation on the BEV detection and segmentation task and showcase its capability in safety-critical data generation.

Despite recent advancements in high-fidelity human reconstruction techniques, the requirements for densely captured images or time-consuming per-instance optimization significantly hinder their applications in broader scenarios. To tackle these issues, we present **HumanSplat**, which predicts the 3D Gaussian Splatting properties of any human from a single input image in a generalizable manner. Specifically, HumanSplat comprises a 2D multi-view diffusion model and a latent reconstruction Transformer with human structure priors that adeptly integrate geometric priors and semantic features within a unified framework. A hierarchical loss that incorporates human semantic information is devised to achieve high-fidelity texture modeling and impose stronger constraints on the estimated multiple views. Comprehensive experiments on standard benchmarks and in-the-wild images demonstrate that HumanSplat surpasses existing state-of-the-art methods in achieving photorealistic novel-view synthesis. Project page: https://humansplat.github.io.

Ensuring robust 3D object detection and localization is crucial for many applications in robotics and autonomous driving. Recent models, however, face difficulties in maintaining high performance when applied to domains with differing sensor setups or geographic locations, often resulting in poor localization accuracy due to domain shift. To overcome this challenge, we introduce a novel diffusion-based box refinement approach. This method employs a domain-agnostic diffusion model, conditioned on the LiDAR points surrounding a coarse bounding box, to simultaneously refine the box's location, size, and orientation. We evaluate this approach under various domain adaptation settings, and our results reveal significant improvements across different datasets, object classes and detectors. Our PyTorch implementation is available at https://github.com/cxy1997/DiffuBox.

Cinematographers adeptly capture the essence of the world, crafting compelling visual narratives through intricate camera movements. Witnessing the strides made by large language models in perceiving and interacting with the 3D world, this study explores their capability to control cameras with human language guidance. We introduce ChatCam, a system that navigates camera movements through conversations with users, mimicking a professional cinematographer's workflow. To achieve this, we propose CineGPT, a GPT-based autoregressive model for text-conditioned camera trajectory generation. We also develop an Anchor Determinator to ensure precise camera trajectory placement. ChatCam understands user requests and employs our proposed tools to generate trajectories, which can be used to render high-quality video footage on radiance field representations. Our experiments, including comparisons to state-of-the-art approaches and user studies, demonstrate our approach's ability to interpret and execute complex instructions for camera operation, showing promising applications in real-world production settings. Project page: https://xinhangliu.com/chatcam.

The ability to promptly respond to environmental changes is crucial for the perception system of autonomous driving. Recently, a new task called streaming perception was proposed. It jointly evaluate the latency and accuracy into a single metric for video online perception. In this work, we introduce StreamDSGN, the first real-time stereo-based 3D object detection framework designed for streaming perception. StreamDSGN is an end-to-end framework that directly predicts the 3D properties of objects in the next moment by leveraging historical information, thereby alleviating the accuracy degradation of streaming perception. Further, StreamDSGN applies three strategies to enhance the perception accuracy: (1) A feature-flow-based fusion method, which generates a pseudo-next feature at the current moment to address the misalignment issue between feature and ground truth. (2) An extra regression loss for explicit supervision of object motion consistency in consecutive frames. (3) A large kernel backbone with a large receptive field for effectively capturing long-range spatial contextual features caused by changes in object positions. Experiments on the KITTI Tracking dataset show that, compared with the strong baseline, StreamDSGN significantly improves the streaming average precision by up to 4.33%. Our code is available at https://github.com/weiyangdaren/streamDSGN-pytorch.

*Do we fully leverage the potential of visual encoder in Multimodal Large Language Models (MLLMs)?* The recent outstanding performance of MLLMs in multimodal understanding has garnered broad attention from both academia and industry. In the current MLLM rat race, the focus seems to be predominantly on the linguistic side. We witness the rise of larger and higher-quality instruction datasets, as well as the involvement of larger-sized LLMs. Yet, scant attention has been directed towards the visual signals utilized by MLLMs, often assumed to be the final high-level features extracted by a frozen visual encoder. In this paper, we introduce the **Dense Connector** - a simple, effective, and plug-and-play vision-language connector that significantly enhances existing MLLMs by leveraging multi-layer visual features, with minimal additional computational overhead. Building on this, we also propose the Efficient Dense Connector, which achieves performance comparable to LLaVA-v1.5 with only 25% of the visual tokens. Furthermore, our model, trained solely on images, showcases remarkable zero-shot capabilities in video understanding as well. Experimental results across various vision encoders, image resolutions, training dataset scales, varying sizes of LLMs (2.7B→70B), and diverse architectures of MLLMs (e.g., LLaVA-v1.5, LLaVA-NeXT and Mini-Gemini) validate the versatility and scalability of our approach, achieving state-of-the-art performance across 19 image and video benchmarks. We hope that this work will provide valuable experience and serve as a basic module for future MLLM development. Code is available at https://github.com/HJYao00/DenseConnector.

We present a differentiable representation, DMesh, for general 3D triangular meshes. DMesh considers both the geometry and connectivity information of a mesh. In our design, we first get a set of convex tetrahedra that compactly tessellates the domain based on Weighted Delaunay Triangulation (WDT), and select triangular faces on the tetrahedra to define the final mesh. We formulate probability of faces to exist on the actual surface in a differentiable manner based on the WDT. This enables DMesh to represent meshes of various topology in a differentiable way, and allows us to reconstruct the mesh under various observations, such as point clouds and multi-view images using gradient-based optimization. We publicize the source code and supplementary material at our project page (https://sonsang.github.io/dmesh-project).

3D Gaussian splatting has recently emerged as a promising technique for novel view synthesis from sparse image sets, yet comes at the cost of requiring millions of 3D Gaussian primitives to reconstruct each 3D scene. This largely limits its application to resource-constrained devices and applications. Despite advances in Gaussian pruning techniques that aim to remove individual 3D Gaussian primitives, the significant reduction in primitives often fails to translate into commensurate increases in rendering speed, impeding efficiency and practical deployment. We identify that this discrepancy arises due to the overlooked impact of fragment count per Gaussian (i.e., the number of pixels each Gaussian is projected onto). To bridge this gap and meet the growing demands for efficient on-device 3D Gaussian rendering, we propose fragment pruning, an orthogonal enhancement to existing pruning methods that can significantly accelerate rendering by selectively pruning fragments within each Gaussian. Our pruning framework dynamically optimizes the pruning threshold for each Gaussian, markedly improving rendering speed and quality. Extensive experiments in both static and dynamic scenes validate the effectiveness of our approach. For instance, by integrating our fragment pruning technique with state-of-the-art Gaussian pruning methods, we achieve up to a 1.71$\times$ speedup on an edge GPU device, the Jetson Orin NX, and enhance rendering quality by an average of 0.16 PSNR on the Tanks\&Temples dataset. Our code is available at https://github.com/GATECH-EIC/Fragment-Pruning.