NeurIPS 2024 Papers

Neural 3D scene representations have shown great potential for 3D reconstruction from 2D images. However, reconstructing real-world captures of complex scenes still remains a challenge. Existing generic 3D reconstruction methods often struggle to represent fine geometric details and do not adequately model reflective surfaces of large-scale scenes. Techniques that explicitly focus on reflective surfaces can model complex and detailed reflections by exploiting better reflection parameterizations. However, we observe that these methods are often not robust in real scenarios where non-reflective as well as reflective components are present. In this work, we propose UniSDF, a general purpose 3D reconstruction method that can reconstruct large complex scenes with reflections. We investigate both camera view as well as reflected view-based color parameterization techniques and find that explicitly blending these representations in 3D space enables reconstruction of surfaces that are more geometrically accurate, especially for reflective surfaces. We further combine this representation with a multi-resolution grid backbone that is trained in a coarse-to-fine manner, enabling faster reconstructions than prior methods. Extensive experiments on object-level datasets DTU, Shiny Blender as well as unbounded datasets Mip-NeRF 360 and Ref-NeRF real demonstrate that our method is able to robustly reconstruct complex large-scale scenes with fine details and reflective surfaces, leading to the best overall performance. Project page: https://fangjinhuawang.github.io/UniSDF.

World models can foresee the outcomes of different actions, which is of paramount importance for autonomous driving. Nevertheless, existing driving world models still have limitations in generalization to unseen environments, prediction fidelity of critical details, and action controllability for flexible application. In this paper, we present Vista, a generalizable driving world model with high fidelity and versatile controllability. Based on a systematic diagnosis of existing methods, we introduce several key ingredients to address these limitations. To accurately predict real-world dynamics at high resolution, we propose two novel losses to promote the learning of moving instances and structural information. We also devise an effective latent replacement approach to inject historical frames as priors for coherent long-horizon rollouts. For action controllability, we incorporate a versatile set of controls from high-level intentions (command, goal point) to low-level maneuvers (trajectory, angle, and speed) through an efficient learning strategy. After large-scale training, the capabilities of Vista can seamlessly generalize to different scenarios. Extensive experiments on multiple datasets show that Vista outperforms the most advanced general-purpose video generator in over 70% of comparisons and surpasses the best-performing driving world model by 55% in FID and 27% in FVD. Moreover, for the first time, we utilize the capacity of Vista itself to establish a generalizable reward for real-world action evaluation without accessing the ground truth actions.

Evidential Deep Learning (EDL), grounded in Evidence Theory and Subjective Logic (SL), provides a robust framework to estimate uncertainty for out-of-distribution (OOD) detection alongside traditional classification probabilities.However, the EDL framework is constrained by its focus on evidence that supports only single categories, neglecting the other collective evidences that could corroborate multiple in-distribution categories. This limitation leads to a diminished estimation of uncertainty and a subsequent decline in OOD detection performance.Additionally, EDL encounters the vanishing gradient problem within its fully-connected layers, further degrading classification accuracy.To address these issues, we introduce hyper-domain and propose Hyper-opinion Evidential Deep Learning (HEDL). HEDL extends the evidence modeling paradigm by explicitly integrating sharp evidence, which supports a singular category, with vague evidence that accommodates multiple potential categories.Additionally, we propose a novel opinion projection mechanism that translates hyper-opinion into multinomial-opinion, which is then optimized within the EDL framework to ensure precise classification and refined uncertainty estimation.HEDL integrates evidences across various categories to yield a holistic evidentiary foundation for achieving superior OOD detection. Furthermore, our proposed opinion projection method effectively mitigates the vanishing gradient issue, ensuring classification accuracy without additional model complexity. Extensive experiments over many datasets demonstrate our proposed method outperforms existing OOD detection methods.

While numerous Video Violence Detection (VVD) methods have focused on representation learning in Euclidean space, they struggle to learn sufficiently discriminative features, leading to weaknesses in recognizing normal events that are visually similar to violent events (i.e., ambiguous violence). In contrast, hyperbolic representation learning, renowned for its ability to model hierarchical and complex relationships between events, has the potential to amplify the discrimination between visually similar events. Inspired by these, we develop a novel Dual-Space Representation Learning (DSRL) method for weakly supervised VVD to utilize the strength of both Euclidean and hyperbolic geometries, capturing the visual features of events while also exploring the intrinsic relations between events, thereby enhancing the discriminative capacity of the features. DSRL employs a novel information aggregation strategy to progressively learn event context in hyperbolic spaces, which selects aggregation nodes through layer-sensitive hyperbolic association degrees constrained by hyperbolic Dirichlet energy. Furthermore, DSRL attempts to break the cyber-balkanization of different spaces, utilizing cross-space attention to facilitate information interactions between Euclidean and hyperbolic space to capture better discriminative features for final violence detection. Comprehensive experiments demonstrate the effectiveness of our proposed DSRL.

The pre-trained text-to-image diffusion models have been increasingly employed to tackle the real-world image super-resolution (Real-ISR) problem due to their powerful generative image priors. Most of the existing methods start from random noise to reconstruct the high-quality (HQ) image under the guidance of the given low-quality (LQ) image. While promising results have been achieved, such Real-ISR methods require multiple diffusion steps to reproduce the HQ image, increasing the computational cost. Meanwhile, the random noise introduces uncertainty in the output, which is unfriendly to image restoration tasks. To address these issues, we propose a one-step effective diffusion network, namely OSEDiff, for the Real-ISR problem. We argue that the LQ image contains rich information to restore its HQ counterpart, and hence the given LQ image can be directly taken as the starting point for diffusion, eliminating the uncertainty introduced by random noise sampling. We finetune the pre-trained diffusion network with trainable layers to adapt it to complex image degradations. To ensure that the one-step diffusion model could yield HQ Real-ISR output, we apply variational score distillation in the latent space to conduct KL-divergence regularization. As a result, our OSEDiff model can efficiently and effectively generate HQ images in just one diffusion step. Our experiments demonstrate that OSEDiff achieves comparable or even better Real-ISR results, in terms of both objective metrics and subjective evaluations, than previous diffusion model-based Real-ISR methods that require dozens or hundreds of steps. The source codes are released at https://github.com/cswry/OSEDiff.

Recent studies on generalizable object detection have attracted increasing attention with additional weak supervision from large-scale datasets with image-level labels. However, weakly-supervised detection learning often suffers from image-to-box label mismatch, i.e., image-level labels do not convey precise object information. We design Language Hierarchical Self-training (LHST) that introduces language hierarchy into weakly-supervised detector training for learning more generalizable detectors. LHST expands the image-level labels with language hierarchy and enables co-regularization between the expanded labels and self-training. Specifically, the expanded labels regularize self-training by providing richer supervision and mitigating the image-to-box label mismatch, while self-training allows assessing and selecting the expanded labels according to the predicted reliability. In addition, we design language hierarchical prompt generation that introduces language hierarchy into prompt generation which helps bridge the vocabulary gaps between training and testing. Extensive experiments show that the proposed techniques achieve superior generalization performance consistently across 14 widely studied object detection datasets.

Inverse rendering methods have achieved remarkable performance in reconstructing high-fidelity 3D objects with disentangled geometries, materials, and environmental light. However, they still face huge challenges in reflective surface reconstruction. Although recent methods model the light trace to learn specularity, the ignorance of indirect illumination makes it hard to handle inter-reflections among multiple smooth objects. In this work, we propose Ref-MC2 that introduces the multi-time Monte Carlo sampling which comprehensively computes the environmental illumination and meanwhile considers the reflective light from object surfaces. To address the computation challenge as the times of Monte Carlo sampling grow, we propose a specularity-adaptive sampling strategy, significantly reducing the computational complexity. Besides the computational resource, higher geometry accuracy is also required because geometric errors accumulate multiple times. Therefore, we further introduce a reflection-aware surface model to initialize the geometry and refine it during inverse rendering. We construct a challenging dataset containing scenes with multiple objects and inter-reflections. Experiments show that our method outperforms other inverse rendering methods on various object groups. We also show downstream applications, e.g., relighting and material editing, to illustrate the disentanglement ability of our method.

Given the complexities inherent in visual scenes, such as object occlusion, a comprehensive understanding often requires observation from multiple viewpoints. Existing multi-viewpoint object-centric learning methods typically employ random or sequential viewpoint selection strategies. While applicable across various scenes, these strategies may not always be ideal, as certain scenes could benefit more from specific viewpoints. To address this limitation, we propose a novel active viewpoint selection strategy. This strategy predicts images from unknown viewpoints based on information from observation images for each scene. It then compares the object-centric representations extracted from both viewpoints and selects the unknown viewpoint with the largest disparity, indicating the greatest gain in information, as the next observation viewpoint. Through experiments on various datasets, we demonstrate the effectiveness of our active viewpoint selection strategy, significantly enhancing segmentation and reconstruction performance compared to random viewpoint selection. Moreover, our method can accurately predict images from unknown viewpoints.

Few-shot 3D point cloud semantic segmentation aims to segment query point clouds with only a few annotated support point clouds. Existing prototype-based methods learn prototypes from the 3D support set to guide the segmentation of query point clouds. However, they encounter the challenge of low prototype quality due to constrained semantic information in the 3D support set and class information bias between support and query sets. To address these issues, in this paper, we propose a novel framework called Generated and Pseudo Content guided Prototype Refinement (GPCPR), which explicitly leverages LLM-generated content and reliable query context to enhance prototype quality. GPCPR achieves prototype refinement through two core components: LLM-driven Generated Content-guided Prototype Refinement (GCPR) and Pseudo Query Context-guided Prototype Refinement (PCPR). Specifically, GCPR integrates diverse and differentiated class descriptions generated by large language models to enrich prototypes with comprehensive semantic knowledge. PCPR further aggregates reliable class-specific pseudo-query context to mitigate class information bias and generate more suitable query-specific prototypes. Furthermore, we introduce a dual-distillation regularization term, enabling knowledge transfer between early-stage entities (prototypes or pseudo predictions) and their deeper counterparts to enhance refinement. Extensive experiments demonstrate the superiority of our method, surpassing the state-of-the-art methods by up to 12.10% and 13.75% mIoU on S3DIS and ScanNet, respectively.

Photographs captured in unstructured tourist environments frequently exhibit variable appearances and transient occlusions, challenging accurate scene reconstruction and inducing artifacts in novel view synthesis. Although prior approaches have integrated the Neural Radiance Field (NeRF) with additional learnable modules to handle the dynamic appearances and eliminate transient objects, their extensive training demands and slow rendering speeds limit practical deployments. Recently, 3D Gaussian Splatting (3DGS) has emerged as a promising alternative to NeRF, offering superior training and inference efficiency along with better rendering quality. This paper presents \textit{Wild-GS}, an innovative adaptation of 3DGS optimized for unconstrained photo collections while preserving its efficiency benefits. \textit{Wild-GS} determines the appearance of each 3D Gaussian by their inherent material attributes, global illumination and camera properties per image, and point-level local variance of reflectance. Unlike previous methods that model reference features in image space, \textit{Wild-GS} explicitly aligns the pixel appearance features to the corresponding local Gaussians by sampling the triplane extracted from the reference image. This novel design effectively transfers the high-frequency detailed appearance of the reference view to 3D space and significantly expedites the training process. Furthermore, 2D visibility maps and depth regularization are leveraged to mitigate the transient effects and constrain the geometry, respectively. Extensive experiments demonstrate that \textit{Wild-GS} achieves state-of-the-art rendering performance and the highest efficiency in both training and inference among all the existing techniques. The code can be accessed via: https://github.com/XuJiacong/Wild-GS

Prevalent text-to-video retrieval methods represent multimodal text-video data in a joint embedding space, aiming at bridging the relevant text-video pairs and pulling away irrelevant ones. One main challenge in state-of-the-art retrieval methods lies in the modality gap, which stems from the substantial disparities between text and video and can persist in the joint space. In this work, we leverage the potential of Diffusion models to address the text-video modality gap by progressively aligning text and video embeddings in a unified space. However, we identify two key limitations of existing Diffusion models in retrieval tasks: The L2 loss does not fit the ranking problem inherent in text-video retrieval, and the generation quality heavily depends on the varied initial point drawn from the isotropic Gaussian, causing inaccurate retrieval. To this end, we introduce a new Diffusion-Inspired Truncated Sampler (DITS) that jointly performs progressive alignment and modality gap modeling in the joint embedding space. The key innovation of DITS is to leverage the inherent proximity of text and video embeddings, defining a truncated diffusion flow from the fixed text embedding to the video embedding, enhancing controllability compared to adopting the isotropic Gaussian. Moreover, DITS adopts the contrastive loss to jointly consider the relevant and irrelevant pairs, not only facilitating alignment but also yielding a discriminatively structured embedding. Experiments on five benchmark datasets suggest the state-of-the-art performance of DITS. We empirically find that DITS can also improve the structure of the CLIP embedding space. Code is available at https://github.com/Jiamian- Wang/DITS-text-video-retrieval

Existing vision-language models (VLMs) mostly rely on vision encoders to extract visual features followed by large language models (LLMs) for visual-language tasks. However, the vision encoders set a strong inductive bias in abstracting visual representation, e.g., resolution, aspect ratio, and semantic priors, which could impede the flexibility and efficiency of the VLMs. Training pure VLMs that accept the seamless vision and language inputs, i.e., without vision encoders, remains challenging and rarely explored. Empirical observations reveal that direct training without encoders results in slow convergence and large performance gaps. In this work, we bridge the gap between encoder-based and encoder-free models, and present a simple yet effective training recipe towards pure VLMs. Specifically, we unveil the key aspects of training encoder-free VLMs efficiently via thorough experiments: (1) Bridging vision-language representation inside one unified decoder; (2) Enhancing visual recognition capability via extra supervision. With these strategies, we launch EVE, an encoder-free vision-language model that can be trained and forwarded efficiently. Notably, solely utilizing 35M publicly accessible data, EVE can impressively rival the encoder-based VLMs of similar capacities across multiple vision-language benchmarks. It significantly outperforms the counterpart Fuyu-8B with mysterious training procedures and undisclosed training data. We believe that EVE provides a transparent and efficient route for developing pure decoder-only architecture across modalities.

Large language and vision models have been leading a revolution in visual computing. By greatly scaling up sizes of data and model parameters, the large models learn deep priors which lead to remarkable performance in various tasks. In this work, we present deep prior assembly, a novel framework that assembles diverse deep priors from large models for scene reconstruction from single images in a zero-shot manner. We show that this challenging task can be done without extra knowledge but just simply generalizing one deep prior in one sub-task. To this end, we introduce novel methods related to poses, scales, and occlusion parsing which are keys to enable deep priors to work together in a robust way. Deep prior assembly does not require any 3D or 2D data-driven training in the task and demonstrates superior performance in generalizing priors to open-world scenes. We conduct evaluations on various datasets, and report analysis, numerical and visual comparisons with the latest methods to show our superiority. Project page: https://junshengzhou.github.io/DeepPriorAssembly.

Recently, Gaussian Splatting, a method that represents a 3D scene as a collection of Gaussian distributions, has gained significant attention in addressing the task of novel view synthesis. In this paper, we highlight a fundamental limitation of Gaussian Splatting: its inability to accurately render discontinuities and boundaries in images due to the continuous nature of Gaussian distributions. To address this issue, we propose a novel framework enabling Gaussian Splatting to perform discontinuity-aware image rendering. Additionally, we introduce a B\'ezier-boundary gradient approximation strategy within our framework to keep the ``differentiability'' of the proposed discontinuity-aware rendering process. Extensive experiments demonstrate the efficacy of our framework.

Domain generalization (DG) methods aim to maintain good performance in an unseen target domain by using training data from multiple source domains. While success on certain occasions are observed, enhancing the baseline across most scenarios remains challenging. This work introduces a simple yet effective framework, dubbed learning from multiple experts (LFME), that aims to make the target model an expert in all source domains to improve DG. Specifically, besides learning the target model used in inference, LFME will also train multiple experts specialized in different domains, whose output probabilities provide professional guidance by simply regularizing the logit of the target model. Delving deep into the framework, we reveal that the introduced logit regularization term implicitly provides effects of enabling the target model to harness more information, and mining hard samples from the experts during training. Extensive experiments on benchmarks from different DG tasks demonstrate that LFME is consistently beneficial to the baseline and can achieve comparable performance to existing arts. Code is available at https://github.com/liangchen527/LFME.

This paper studies the problem of estimating physical properties (system identification) through visual observations. To facilitate geometry-aware guidance in physical property estimation, we introduce a novel hybrid framework that leverages 3D Gaussian representation to not only capture explicit shapes but also enable the simulated continuum to render object masks as 2D shape surrogates during training. We propose a new dynamic 3D Gaussian framework based on motion factorization to recover the object as 3D Gaussian point sets across different time states. Furthermore, we develop a coarse-to-fine filling strategy to generate the density fields of the object from the Gaussian reconstruction, allowing for the extraction of object continuums along with their surfaces and the integration of Gaussian attributes into these continuum. In addition to the extracted object surfaces, the Gaussian-informed continuum also enables the rendering of object masks during simulations, serving as 2D-shape guidance for physical property estimation. Extensive experimental evaluations demonstrate that our pipeline achieves state-of-the-art performance across multiple benchmarks and metrics. Additionally, we illustrate the effectiveness of the proposed method through real-world demonstrations, showcasing its practical utility. Our project page is at https://jukgei.github.io/project/gic.

Centred on content modification and style preservation, Scene Text Editing (STE) remains a challenging task despite considerable progress in text-to-image synthesis and text-driven image manipulation recently. GAN-based STE methods generally encounter a common issue of model generalization, while Diffusion-based STE methods suffer from undesired style deviations. To address these problems, we propose TextCtrl, a diffusion-based method that edits text with prior guidance control. Our method consists of two key components: (i) By constructing fine-grained text style disentanglement and robust text glyph structure representation, TextCtrl explicitly incorporates Style-Structure guidance into model design and network training, significantly improving text style consistency and rendering accuracy. (ii) To further leverage the style prior, a Glyph-adaptive Mutual Self-attention mechanism is proposed which deconstructs the implicit fine-grained features of the source image to enhance style consistency and vision quality during inference. Furthermore, to fill the vacancy of the real-world STE evaluation benchmark, we create the first real-world image-pair dataset termed ScenePair for fair comparisons. Experiments demonstrate the effectiveness of TextCtrl compared with previous methods concerning both style fidelity and text accuracy. Project page: https://github.com/weichaozeng/TextCtrl.

We propose ID-to-3D, a method to generate identity- and text-guided 3D human heads with disentangled expressions, starting from even a single casually captured ‘in-the-wild’ image of a subject. The foundation of our approach is anchored in compositionality, alongside the use of task-specific 2D diffusion models as priors for optimization. First, we extend a foundational model with a lightweight expression-aware and ID-aware architecture, and create 2D priors for geometric and texture generation, via fine-tuning only 0.2% of its available training parameters. Then, we jointly leverage a neural parametric representation for the expression of each subject and a multi-stage generation of highly detailed geometry and albedo texture. This combination of strong face identity embeddings and our neural representation enables accurate reconstruction of not only facial features but also accessories and hair, and can be meshed to provide render-ready assets for gaming and telepresence. Our results achieve an unprecedented level of id-consistent and high-quality texture and geometry generation, generalizing to a ‘world’ of unseen 3D identities, without relying on large 3D captured datasets of human assets.

Deep neural networks (DNNs) often suffer from the overconfidence issue, where incorrect predictions are made with high confidence scores, hindering the applications in critical systems. In this paper, we propose a novel approach called Typicalness-Aware Learning (TAL) to address this issue and improve failure detection performance. We observe that, with the cross-entropy loss, model predictions are optimized to align with the corresponding labels via increasing logit magnitude or refining logit direction. However, regarding atypical samples, the image content and their labels may exhibit disparities. This discrepancy can lead to overfitting on atypical samples, ultimately resulting in the overconfidence issue that we aim to address. To address this issue, we have devised a metric that quantifies the typicalness of each sample, enabling the dynamic adjustment of the logit magnitude during the training process. By allowing relatively atypical samples to be adequately fitted while preserving reliable logit direction, the problem of overconfidence can be mitigated. TAL has been extensively evaluated on benchmark datasets, and the results demonstrate its superiority over existing failure detection methods. Specifically, TAL achieves a more than 5\% improvement on CIFAR100 in terms of the Area Under the Risk-Coverage Curve (AURC) compared to the state-of-the-art. Code is available at https://github.com/liuyijungoon/TAL.