CVPR 2025 Papers

Pre-trained vision models (PVMs) are fundamental to modern robotics, yet their optimal configuration remains unclear. Through systematic evaluation, we find that while DINO and iBOT outperform MAE across visuomotor control and perception tasks, they struggle when trained on non-(single-)object-centric (NOC) data--a limitation strongly correlated with their diminished ability to learn object-centric representations. This investigation indicates that the ability to form object-centric representations from the non-object-centric robotics dataset is the key to success for PVMs. Motivated by this discovery, we designed SlotMIM, a method that induces object-centric representations by introducing a semantic bottleneck to reduce the number of prototypes to encourage the emergence of objectness as well as cross-view consistency regularization for encouraging multiview invariance. Our experiments encompass pre-training on object-centric, scene-centric, web-crawled, and ego-centric data. Across all settings, our approach learns transferrable representations and achieves significant improvements over prior work in image recognition, scene understanding, and robot learning evaluations. When scaled up with million-scale datasets, our method also demonstrates superior data efficiency and scalability. Our code and models are publicly available at https://github.com/CVMI-Lab/SlotMIM.

Both text-to-image generation and large language models (LLMs) have made significant advancements. However, many text-to-image models still employ the somewhat outdated T5 and CLIP as their text encoders. In this work, we investigate the effectiveness of using modern decoder-only LLMs as text encoders for text-to-image diffusion models. We build a standardized training and evaluation pipeline that allows us to isolate and evaluate the effect of different text embeddings. We train a total of 27 text-to-image models with 12 different text encoders to analyze the critical aspects of LLMs that could impact text-to-image generation, including the approaches to extract embeddings, different LLMs variants, and model sizes. Our experiments reveal that the de facto way of using last-layer embeddings as conditioning leads to inferior performance. Instead, we explore embeddings from various layers and find that using layer-normalized averaging across all layers significantly improves alignment with complex prompts. Most LLMs with this conditioning outperform the baseline T5 model, showing enhanced performance in advanced visio-linguistic reasoning skills.

Successful forensic detectors can produce excellent results in supervised learning benchmarks but struggle to transfer to real-world applications. We believe this limitation is largely due to inadequate training data quality. While most research focuses on developing new algorithms, less attention is given to training data selection, despite evidence that performance can be strongly impacted by spurious correlations such as content, format, or resolution. A well-designed forensic detector should detect generator specific artifacts rather than reflect data biases. To this end, we propose B-Free, a bias-free training paradigm, where fake images are generated from real ones using the conditioning procedure of stable diffusion models. This ensures semantic alignment between real and fake images, allowing any differences to stem solely from the subtle artifacts introduced by AI generation. Through content-based augmentation, we show significant improvements in both generalization and robustness over state-of-the-art detectors and more calibrated results across 27 different generative models, including recent releases, like FLUX and Stable Diffusion 3.5. Our findings emphasize the importance of a careful dataset design, highlighting the need for further research on this topic. Code and data are publicly available at https://grip-unina.github.io/B-Free/.

Pre-training & fine-tuning can enhance the transferring efficiency and performance in visual tasks. Recent delta-tuning methods provide more options for visual classification tasks. Despite their success, existing visual delta-tuning art fails to exceed the upper limit of full fine-tuning on challenging tasks. To find a competitive alternative to full fine-tuning, we propose the Multi-cognitive Visual Adapter (Mona) tuning, a novel adapter-based tuning method. First, we introduce multiple vision-friendly filters into the adapter to enhance its ability for processing visual signals, while previous methods mainly rely on language-friendly linear filters. Second, we add the scaled layernorm in the adapter to regulate the distribution of input features for visual filters. To fully demonstrate the practicality and generality of Mona, we conduct experiments on representative visual tasks, including instance segmentation on COCO, semantic segmentation on ADE20K, object detection on Pascal VOC, oriented object detection on DOTA/STAR, and image classification on three common datasets. Exciting results illustrate that Mona surpasses full fine-tuning on all these tasks by tuning less than 5% params of the backbone, and is the only delta-tuning method outperforming full fine-tuning on all tasks. For example, Mona achieves 1% performance gain on the COCO compared to full fine-tuning. Comprehensive results suggest that Mona-tuning is more suitable for retaining and utilizing the capabilities of pre-trained models than full fine-tuning. The code is publicly available on https://github.com/Leiyi-Hu/mona.

We propose 4Real-Video, a novel framework for generating 4D videos, organized as a grid of video frames with both time and viewpoint axes. In this grid, each row contains frames sharing the same timestep, while each column contains frames from the same viewpoint. One stream performs viewpoint updates on columns, and the other stream performs temporal updates on rows. After each diffusion transformer layer, a newly designed synchronization layer exchanges information between the two token streams. We propose two implementations of the synchronization layer, using either hard or soft synchronization.This feedforward architecture improves upon previous work in three ways: higher inference speed, enhanced visual quality (measured by FVD, CLIP, and VideoScore), and improved temporal and viewpoint consistency (measured by VideoScore, GIM-Confidence, and Dust3R-Confidence).

Generating realistic intermediate shapes between non-rigidly deformed shapes is a challenging task in computer vision, especially with unstructured data (e.g., point clouds) where temporal consistency across frames is lacking, and topologies are changing. Most interpolation methods are designed for structured data (i.e., meshes) and do not apply to real-world point clouds. In contrast, our approach leverages neural implicit representation (NIR) to enable free-topology changing shape deformation. Unlike previous mesh-based methods, which model learns vertex-based deformation fields, our method learns a continuous velocity field in Euclidean space, making it suitable for less structured data such as point clouds.Additionally, our method does not require intermediate-shape supervision during training; instead, we incorporate physical and geometrical constraints to regularize the velocity field. We reconstruct intermediate surfaces using a modified level-set equation, directly linking our NIR with the velocity field. Experiments show that our method significantly outperforms previous NIR approaches across various scenarios (e.g., noisy, partial, topology-changing, non-isometric shapes) and, for the first time, enables new applications like 4D Kinect sequence upsampling and real-world high-resolution mesh deformation.

We propose the first 4D tracking and mapping method that jointly performs camera localization and non-rigid surface reconstruction via differentiable rendering. Our approach captures 4D scenes from an online stream of color images with depth measurements or predictions by simultaneously optimizing scene geometry, appearance, dynamics, and camera ego-motion. Although natural environments exhibit complex non-rigid motions, 4D-SLAM remains relatively underexplored due to its inherent challenges; even with 2.5D signals, the problem is ill-posed because of the high dimensionality of the optimization space. To overcome these challenges, we first introduce a SLAM method based on Gaussian surface primitives that leverages depth signals more effectively than 3D Gaussians, thereby achieving accurate surface reconstruction. To further model non-rigid deformations, we employ a warp-field represented by a multi-layer perceptron (MLP) and introduce a novel camera pose estimation technique along with surface regularization terms that facilitate spatio-temporal reconstruction. In addition to these algorithmic challenges, a significant hurdle in 4D-SLAM research is the lack of publicly available datasets with reliable ground truth and evaluation protocols. To address this, we present a novel synthetic dataset of everyday objects with diverse motions, leveraging large-scale object models and animation modeling. In summary, we open up the modern 4D-SLAM research by introducing a novel method and evaluation protocols grounded in modern vision and rendering techniques.

3D Gaussian Splatting (3DGS) has substantial potential for enabling photorealistic Free-Viewpoint Video (FVV) experiences. However, the vast number of Gaussians and their associated attributes poses significant challenges for storage and transmission. Existing methods typically handle dynamic 3DGS representation and compression separately, neglecting motion information and the rate-distortion (RD) trade-off during training, leading to performance degradation and increased model redundancy. To address this gap, we propose 4DGC, a novel rate-aware 4D Gaussian compression framework that significantly reduces storage size while maintaining superior RD performance for FVV. Specifically, 4DGC introduces a motion-aware dynamic Gaussian representation that utilizes a compact motion grid combined with sparse compensated Gaussians to exploit inter-frame similarities. This representation effectively handles large motions, preserving quality and reducing temporal redundancy. Furthermore, we present an end-to-end compression scheme that employs differentiable quantization and a tiny implicit entropy model to compress the motion grid and compensated Gaussians efficiently. The entire framework is jointly optimized using a rate-distortion trade-off. Extensive experiments demonstrate that 4DGC supports variable bitrates and consistently outperforms existing methods in RD performance across multiple datasets.

Learning 4D language fields to enable time-sensitive, open-ended language queries in dynamic scenes is essential for many real-world applications. While LangSplat successfully grounds CLIP features into 3D Gaussian representations, achieving precision and efficiency in 3D static scenes, it lacks the ability to handle dynamic 4D fields as CLIP, designed for static image-text tasks, cannot capture temporal dynamics in videos. Real-world environments are inherently dynamic, with object semantics evolving over time. Building a precise 4D language field necessitates obtaining pixel-aligned, object-wise video features, which current vision models struggle to achieve. To address these challenges, we propose 4D LangSplat, which learns 4D language fields to handle time-agnostic or time-sensitive open-vocabulary queries in dynamic scenes efficiently. 4D LangSplat bypasses learning the language field from vision features and instead learns directly from text generated from object-wise video captions via Multimodal Large Language Models (MLLMs). Specifically, we propose a multimodal object-wise video prompting method, consisting of visual and text prompts that guide MLLMs to generate detailed, temporally consistent, high-quality captions for objects throughout a video. These captions are encoded using a Large Language Model into high-quality sentence embeddings, which then serve as pixel-aligned, object-specific feature supervision, facilitating open-vocabulary text queries through shared embedding spaces. Recognizing that objects in 4D scenes exhibit smooth transitions across states, we further propose a status deformable network to model these continuous changes over time effectively. Our results across multiple benchmarks demonstrate that 4D LangSplat attains precise and efficient results for both time-sensitive and time-agnostic open-vocabulary queries.

The increasing demand for high-quality 3D assets across various industries necessitates efficient and automated 3D content creation. Despite recent advancements in 3D generative models, existing methods still face challenges with optimization speed, geometric fidelity, and the lack of assets for physically based rendering (PBR). In this paper, we introduce 3DTopia-XL, a scalable native 3D generative model designed to overcome these limitations. 3DTopia-XL leverages a novel primitive-based 3D representation, PrimX, which encodes detailed shape, albedo, and material field into a compact tensorial format, facilitating the modeling of high-resolution geometry with PBR assets. On top of the novel representation, we propose a generative framework based on Diffusion Transformer (DiT), which comprises 1) Primitive Patch Compression, 2) and Latent Primitive Diffusion. 3DTopia-XL learns to generate high-quality 3D assets from textual or visual inputs. Extensive qualitative and quantitative experiments are conducted to demonstrate that 3DTopia-XL significantly outperforms existing methods in generating high-quality 3D assets with fine-grained textures and materials, efficiently bridging the quality gap between generative models and real-world applications.

3D Gaussian Splatting (3DGS) enables efficient reconstruction and high-fidelity real-time rendering of complex scenes on consumer hardware. However, due to its rasterization-based formulation, 3DGS is constrained to ideal pinhole cameras and lacks support for secondary lighting effects. Recent methods address these limitations by tracing the particles instead, but, this comes at the cost of significantly slower rendering. In this work, we propose 3D Gaussian Unscented Transform (3DGUT), replacing the EWA splatting formulation with the Unscented Transform that approximates the particles through sigma points, which can be projected exactly under any nonlinear projection function. This modification enables trivial support of distorted cameras with time dependent effects such as rolling shutter, while retaining the efficiency of rasterization. Additionally, we align our rendering formulation with that of tracing-based methods, enabling secondary ray tracing required to represent phenomena such as reflections and refraction within the same 3D representation. The source code is available at: https://github.com/nv-tlabs/3dgrut.

Despite advances in neural rendering, due to the scarcity of high-quality 3D datasets and the inherent limitations of multi-view diffusion models, view synthesis and 3D model generation are restricted to low resolutions with suboptimal multi-view consistency. In this study, we present a novel 3D enhancement pipeline, dubbed 3DEnhancer, which employs a multi-view latent diffusion model to enhance coarse 3D inputs while preserving multi-view consistency. Our method includes a pose-aware encoder and a diffusion-based denoiser to refine low-quality multi-view images, along with data augmentation and a multi-view attention module with epipolar aggregation to maintain consistent, high-quality 3D outputs across views. Unlike existing video-based approaches, our model supports seamless multi-view enhancement with improved coherence across diverse viewing angles. Extensive evaluations show that 3DEnhancer significantly outperforms existing methods, boosting both multi-view enhancement and per-instance 3D optimization tasks.

We propose 3D-SLNR, a new and ultra-lightweight neural representation with outstanding performance for large-scale 3D mapping. The representation defines a global signed distance function (SDF) in near-surface space based on a set of band-limited local SDFs anchored at support points sampled from point clouds. These SDFs are parameterized only by a tiny multi-layer perceptron (MLP) with no latent features, and the state of each SDF is modulated by three learnable geometric properties: position, rotation, and scaling, which make the representation adapt to complex geometries. Then, we develop a novel parallel algorithm tailored for this unordered representation to efficiently detect local SDFs where each sampled point is located, allowing for real-time updates of local SDF states during training. Additionally, a prune-and-expand strategy is introduced to enhance adaptability further. The synergy of our low-parameter model and its adaptive capabilities results in an extremely compact representation with excellent expressiveness. Extensive experiments demonstrate that our method achieves state-of-the-art reconstruction performance with less than 1/5 of the memory footprint compared with previous advanced methods.

Constructing compact and informative 3D scene representations is essential for effective embodied exploration and reasoning, especially in complex environments over extended periods. Existing representations, such as object-centric 3D scene graphs, oversimplify spatial relationships by modeling scenes as isolated objects with restrictive textual relationships, making it difficult to address queries requiring nuanced spatial understanding. Moreover, these representations lack natural mechanisms for active exploration and memory management, hindering their application to lifelong autonomy. In this work, we propose 3D-Mem, a novel 3D scene memory framework for embodied agents. 3D-Mem employs informative multi-view images, termed Memory Snapshots, to capture rich visual information of explored regions. It further integrates frontier-based exploration by introducing Frontier Snapshots--glimpses of unexplored areas--enabling agents to make decisions by considering both known and potential new information. To support lifelong memory in active exploration settings, we present an incremental construction pipeline for 3D-Mem, as well as a memory retrieval technique for memory management. Experimental results on three benchmarks demonstrate that 3D-Mem significantly enhances agents' exploration and reasoning capabilities in 3D environments, highlighting its potential for advancing applications in embodied AI.

Recent works have shown that visual pretraining on egocentric datasets using masked autoencoders (MAE) can improve generalization for downstream robotics tasks. However, these approaches pretrain only on 2D images, while many robotics applications require 3D scene understanding. In this work, we propose 3D-MVP, a novel approach for 3D Multi-View Pretraining using masked autoencoders. We leverage Robotic View Transformer (RVT), which uses a multi-view transformer to understand the 3D scene and predict gripper pose actions. We split RVT's multi-view transformer into visual encoder and action decoder, and pretrain its visual encoder using masked autoencoding on large-scale 3D datasets such as Objaverse. We evaluate 3D-MVP on a suite of virtual robot manipulation tasks and demonstrate improved performance over baselines. Our results suggest that 3D-aware pretraining is a promising approach to improve generalization of vision-based robotic manipulation policies.

Current 3D Large Multimodal Models (3D LMMs) have shown tremendous potential in 3D-vision-based dialogue and reasoning. However, how to further enhance 3D LMMs to achieve fine-grained scene understanding and facilitate flexible human-agent interaction remains a challenging problem. In this work, we introduce 3D-LLaVA, a simple yet highly powerful 3D LMM designed to act as an intelligent assistant in comprehending, reasoning, and interacting with the 3D world. Unlike existing top-performing methods that rely on complicated pipelines--such as offline multi-view feature extraction or additional task-specific heads--3D-LLaVA adopts a minimalist design with integrated architecture and only takes point clouds as input. At the core of 3D-LLaVA is a new Omni Superpoint Transformer (OST), which integrates three functionalities: (1) a visual feature selector that converts and selects visual tokens, (2) a visual prompt encoder that embeds interactive visual prompts into the visual token space, and (3) a referring mask decoder that produces 3D masks based on text description. This versatile OST is empowered by the hybrid pretraining to obtain perception priors and leveraged as the visual connector that bridges the 3D data to the LLM. After performing unified instruction tuning, our 3D-LLaVA reports impressive results on various benchmarks. The code and model will be released at https://github.com/djiajunustc/3D-LLaVA.

Photo-realistic image rendering from 3D scene reconstruction has advanced significantly with neural rendering techniques. Among these, 3D Gaussian Splatting (3D-GS) outperforms Neural Radiance Fields (NeRFs) in quality and speed but struggles with shape and color discontinuities. We propose 3D Half-Gaussian (3D-HGS) kernels as a plug-and-play solution to address these limitations. Our experiments show that 3D-HGS enhances existing 3D-GS methods, achieving state-of-the-art rendering quality without compromising speed. More demos and code are available at https://lihaolin88.github.io/CVPR-2025-3DHGS