CVPR 2025 Papers

Reconstructing the geometry and appearance of objects from photographs taken in different environments is difficult as the illumination and therefore the object appearance vary across captured images. This is particularly challenging for more specular objects whose appearance strongly depends on the viewing direction. Some prior approaches model appearance variation across images using a per-image embedding vector, while others use physically-based rendering to recover the materials and per-image illumination. Such approaches fail at faithfully recovering view-dependent appearance given significant variation in input illumination and tend to produce mostly diffuse results. We present an approach that reconstructs objects from images taken under different illuminations by first relighting the images under a single reference illumination with a multiview relighting diffusion model and then reconstructing the object's geometry and appearance with a radiance field architecture that is robust to the small remaining inconsistencies among the relit images. We validate our proposed approach on both simulated and real datasets and demonstrate that it greatly outperforms existing techniques at reconstructing high-fidelity appearance from images taken under extreme illumination variation. Moreover, our approach is particularly effective at recovering view-dependent "shiny" appearance which cannot be reconstructed by prior methods.

Recent endeavors in Multimodal Large Language Models (MLLMs) aim to unify visual comprehension and generation by combining LLM and diffusion models, the state-of-the-art in each task, respectively. Existing approaches rely on spatial visual tokens, where image patches are encoded and arranged according to a spatial order (e.g., raster scan). However, we show that spatial tokens lack the recursive structure inherent to languages, hence form an impossible language for LLM to master. In this paper, we build a proper visual language by leveraging diffusion timesteps to learn discrete, recursive visual tokens. Our proposed tokens recursively compensate for the progressive attribute loss in noisy images as timesteps increase, enabling the diffusion model to reconstruct the original image at any timestep. This approach allows us to effectively integrate the strengths of LLMs in autoregressive reasoning and diffusion models in precise image generation, achieving seamless multimodal comprehension and generation within a unified framework. Extensive experiments show that we achieve a new SOTA for multimodal comprehension and generation simultaneously compared with other MLLMs. Project Page: https://DDT-LLaMA.github.io/.

Multi-modal understanding plays a crucial role in artificial intelligence by enabling models to jointly interpret inputs from different modalities. However, conventional approaches such as contrastive learning often struggle with modality discrepancies, leading to potential misalignments. In this paper, we propose a novel class anchor alignment approach that leverages class probability distributions for multi-modal representation learning. Our method, Class-anchor-ALigned generative Modeling (CALM), encodes class anchors as prompts to generate and align class probability distributions for each modality, enabling more flexible alignment. Furthermore, we introduce a cross-modal probabilistic variational autoencoder to model uncertainty in the alignment, enhancing the ability to capture deeper relationships between modalities and data variations. Extensive experiments on four benchmark datasets demonstrate that our approach significantly outperforms state-of-the-art methods, especially in out-of-domain evaluations. This highlights its superior generalization capabilities in multi-modal representation learning.

Generative inbetweening aims to generate intermediate frame sequences by utilizing two key frames as input. Although remarkable progress has been made in video generation models, generative inbetweening still faces challenges in maintaining temporal stability due to the ambiguous interpolation path between two key frames. This issue becomes particularly severe when there is a large motion gap between input frames. In this paper, we propose a straightforward yet highly effective Frame-wise Conditions-driven Video Generation (FCVG) method that significantly enhances the temporal stability of interpolated video frames. Specifically, our FCVG provides an explicit condition for each frame, making it much easier to identify the interpolation path between two input frames and thus ensuring temporally stable production of visually plausible video frames. To achieve this, we suggest extracting matched lines from two input frames that can then be easily interpolated frame by frame, serving as frame-wise conditions seamlessly integrated into existing video generation models. In extensive evaluations covering diverse scenarios such as natural landscapes, complex human poses, camera movements and animations, existing methods often exhibit incoherent transitions across frames. In contrast, our FCVG demonstrates the capability to generate temporally stable videos using both linear and non-linear interpolation curves. Our project page and code are available at https://fcvg-inbetween.github.io/.

Recent advancements in large generative models, particularly diffusion-based methods, have significantly enhanced the capabilities of image editing. However, achieving precise control over image composition tasks remains a challenge. Layered representations, which allow for independent editing of image components, are essential for user-driven content creation, yet existing approaches often struggle to decompose image into plausible layers with accurately retained transparent visual effects such as shadows and reflections. We propose LayerDecomp, a generative framework for image layer decomposition which outputs photorealistic clean backgrounds and high-quality transparent foregrounds with faithfully preserved visual effects. To enable effective training, we first introduce a dataset preparation pipeline that automatically scales up simulated multi-layer data with synthesized visual effects. To further enhance real-world applicability, we supplement this simulated dataset with camera-captured images containing natural visual effects. Additionally, we propose a consistency loss which enforces the model to learn accurate representations for the transparent foreground layer when ground-truth annotations are not available. Our method achieves superior quality in layer decomposition, outperforming existing approaches in object removal and spatial editing tasks across several benchmarks and multiple user studies, unlocking various creative possibilities for layer-wise image editing.

The recent progress in semantic point cloud segmentation is attributed to deep networks, which require a large amount of point cloud data for training. However, how to collect substantial point-wise annotations of the point clouds at affordable cost for the end-to-end network training still needs to be solved. In this paper, we propose Generative Hard Example Augmentation (GHEA) to achieve novel examples of point clouds, which enrich the data for training the segmentation network. Firstly, GHEA employs the generative network to embed the discrepancy between the point clouds into the latent space. From the latent space, we sample multiple discrepancies for reshaping a point cloud to various examples, contributing to the richness of the training data. Secondly, GHEA mixes the reshaped point clouds by respecting their segmentation errors. This mixup allows the reshaped point clouds, which are difficult to segment, to join as the challenging example for network training. We evaluate the effectiveness of GHEA, which helps the popular segmentation networks to improve the performances.

3D city generation with NeRF-based methods shows promising generation results but is computationally inefficient. Recently 3D Gaussian splatting (3D-GS) has emerged as a highly efficient alternative for object-level 3D generation. However, adapting 3D-GS from finite-scale 3D objects and humans to infinite-scale 3D cities is non-trivial. Unbounded 3D city generation entails significant storage overhead (out-of-memory issues), arising from the need to expand points to billions, often demanding hundreds of Gigabytes of VRAM for a city scene spanning 10km^2. In this paper, we propose **GaussianCity**, a generative Gaussian splatting framework dedicated to efficiently synthesizing unbounded 3D cities with a single feed-forward pass. Our key insights are two-fold: **1)** *Compact 3D Scene Representation*: We introduce BEV-Point as a highly compact intermediate representation, ensuring that the growth in VRAM usage for unbounded scenes remains constant, thus enabling unbounded city generation. **2)** *Spatial-aware Gaussian Attribute Decoder*: We present spatial-aware BEV-Point decoder to produce 3D Gaussian attributes, which leverages Point Serializer to integrate the structural and contextual characteristics of BEV points. Extensive experiments demonstrate that GaussianCity achieves state-of-the-art results in both drone-view and street-view 3D city generation. Notably, compared to CityDreamer, GaussianCity exhibits superior performance with a speedup of 60 times (10.72 FPS v.s. 0.18 FPS).

Generalized feed-forward Gaussian models have made significant strides in sparse-view 3D reconstruction by leveraging prior knowledge from large multi-view datasets. However, these models often struggle to represent high-frequency details primarily due to the limited number of Gaussians. While the densification strategy from per-scene 3D Gaussian splatting can be adapted for feed-forward models, it typically requires tens of thousands of optimization steps to reconstruct fine details and can easily lead to overfitting in sparse-view scenarios. In this paper, we propose Generative Densification, an efficient and generalizable densification strategy specifically tailored for feed-forward models. Instead of iteratively splitting and cloning raw Gaussian parameters, our method up-samples feature representations produced by feed-forward models and generates their corresponding fine Gaussians in a single forward pass, leveraging learned prior knowledge for enhanced generalization. Experimental results on both object-level and scene-level reconstruction tasks demonstrate that our method outperforms state-of-the-art approaches with comparable or smaller model sizes, achieving notable improvements in representing fine details.

Generalizable object keypoint localization is a fundamental computer vision task in understanding the object structure. It is challenging for existing keypoint localization methods because their limited training data cannot provide generalizable shape and semantic cues, leading to inferior performance and generalization capability. Instead of relying on large scale training data, this work tackles this challenge by exploiting the rich priors from large generative models. We propose a data-efficient generalizable localization method named GenLoc. GenLoc extracts the generative priors from a pre-trained image generation model by calculating the correlation map between image latent feature and condition embedding. Those priors are hence optimized with our proposed heatmap expectation loss to perform object keypoint localization. Benefited by the rich knowledge of generative priors in understanding of object semantics and structures, GenLoc achieves superior performance on various object keypoint localization benchmarks. It shows more substantial performance enhancements in cross-domain, few-shot and zero-shot evaluation settings, e.g., getting 20%+ AP enhancement over CLAMP in various zero-shot settings.

Existing point cloud completion methods, which typically depend on predefined synthetic training datasets, encounter significant challenges when applied to out-of-distribution, real-world scans. To overcome this limitation, we introduce a zero-shot completion framework, termed GenPC, designed to reconstruct high-quality real-world scans by leveraging explicit 3D generative priors.Our key insight is that recent feed-forward 3D generative models, trained on extensive internet-scale data, have demonstrated the ability to perform 3D generation from single-view images in a zero-shot setting. To harness this for completion, we first develop a Depth Prompting module that links partial point clouds with image-to-3D generative models by leveraging depth images as a stepping stone. To retain the original partial structure in the final results, we designed the Geometric Preserving Fusion module that aligns the generated shape with input by adaptively adjusting its pose and scale.Extensive experiments on widely used benchmarks validate the superiority and generalizability of our approach, bringing us a step closer to robust real-world scan completion.

High-quality 3D assets for traffic participants are critical for multi-sensor simulation, which is essential for the safe end-to-end development of autonomy. Building assets from in-the-wild data is key for diversity and realism, but existing neural-rendering based reconstruction methods are slow and generate assets that render well only from viewpoints close to the original observations, limiting their usefulness in simulation. Recent diffusion-based generative models build complete and diverse assets, but perform poorly on in-the-wild driving scenes, where observed actors are captured under sparse and limited fields of view, and are partially occluded. In this work, we propose a 3D latent diffusion model that learns on in-the-wild LiDAR and camera data captured by a sensor platform and generates high-quality 3D assets with complete geometry and appearance. Key to our method is a "reconstruct-then-generate" approach that first leverages occlusion-aware neural rendering trained over multiple scenes to build a high-quality latent space for objects, and then trains a diffusion model that operates on the latent space. We show our method outperforms existing reconstruction and generation based methods, unlocking diverse and scalable content creation for simulation. Please visit https://waabi.ai/genassets for more details.

Aiming to reduce the computational cost of Softmax in massive label space of Face Recognition (FR) benchmarks, recent studies estimate the output using a subset of identities. Although promising, the association between the computation cost and the number of identities in the dataset remains linear only with a reduced ratio. A shared characteristic among available FR methods is the employment of atomic scalar labels during training. Consequently, the input to label matching is through a dot product between the feature vector of the input and the Softmax centroids. Inspired by generative modeling, we present a simple yet effective method that substitutes scalar labels with structured identity code, i.e., a sequence of integers. Specifically, we propose a tokenization scheme that transforms atomic scalar labels into structured identity codes. Then, we train an FR backbone to predict the code for each input instead of its scalar label. As a result, the associated computational cost becomes logarithmic \wrt number of identities.We demonstrate the benefits of the proposed method by conducting experiments. In particular, our method outperforms its competitors by 1.52%, and 0.6% at TAR@FAR=1e-4 on IJB-B and IJB-C, respectively, while transforming the association between computational cost and the number of identities from linear to logarithmic. \href https://github.com/msed-Ebrahimi/GIF Code

Generative retrieval is an emerging approach in information retrieval that generates identifiers (IDs) of target data based on a query, providing an efficient alternative to traditional embedding-based retrieval methods. However, existing models are task-specific and fall short of embedding-based retrieval in performance. This paper proposes GENIUS, a universal generative retrieval framework supporting diverse tasks across multiple modalities and domains. At its core, GENIUS introduces modality-decoupled semantic quantization, transforming multimodal data into discrete IDs encoding both modality and semantics. Moreover, to enhance generalization, we propose a query augmentation that interpolates between a query and its target, allowing GENIUS to adapt to varied query forms. Evaluated on the M-BEIR benchmark, it surpasses prior generative methods by a clear margin. Unlike embedding-based retrieval, GENIUS consistently maintains high retrieval speed across database size, with competitive performance across multiple benchmarks. With additional re-ranking, GENIUS often achieves results close to those of embedding-based methods while preserving efficiency.

Color constancy methods often struggle to generalize across different camera sensors due to varying spectral sensitivities. We present GCC, which leverages diffusion models to inpaint color checkers into images for illumination estimation. Our key innovations include (1) a single-step deterministic inference approach that inpaints color checkers reflecting scene illumination, (2) a Laplacian decomposition technique that preserves checker structure while allowing illumination-dependent color adaptation, and (3) a mask-based data augmentation strategy for handling imprecise color checker annotations. By harnessing rich priors from pre-trained diffusion models, GCC demonstrates strong robustness in challenging cross-camera scenarios. These results highlight our method's effective generalization capability across different camera characteristics without requiring sensor-specific training, making it a versatile and practical solution for real-world applications.

Recent advances in imitation learning for 3D robotic manipulation have shown promising results with diffusion-based policies. However, achieving human-level dexterity requires seamless integration of geometric precision and semantic understanding. We present G3Flow, a novel framework that constructs real-time semantic flow, a dynamic, object-centric 3D semantic representation by leveraging foundation models. Our approach uniquely combines 3D generative models for digital twin creation, vision foundation models for semantic feature extraction, and robust pose tracking for continuous semantic flow updates. This integration enables complete semantic understanding even under occlusions while eliminating manual annotation requirements. By incorporating semantic flow into diffusion policies, we demonstrate significant improvements in both terminal-constrained manipulation and cross-object generalization. Extensive experiments across five simulation tasks show that G3Flow consistently outperforms existing approaches, achieving up to 68.3% and 50.1% average success rates on terminal-constrained manipulation and cross-object generalization tasks respectively. Our results demonstrate the effectiveness of G3Flow in enhancing real-time dynamic semantic feature understanding for robotic manipulation policies.

Person re-identification (ReID) aims to extract accurate identity representation features. However, during feature extraction, individual samples are inevitably affected by noise (background, occlusions, and model limitations). Considering that features from the same identity follow a normal distribution around identity centers after training, we propose a Training-Free Feature Centralization ReID framework (Pose2ID) by aggregating the same identity features to reduce individual noise and enhance the stability of identity representation, which preserves the feature's original distribution for following strategies such as re-ranking. Specifically, to obtain samples of the same identity, we introduce two components: Identity-Guided Pedestrian Generation: by leveraging identity features to guide the generation process, we obtain high-quality images with diverse poses, ensuring identity consistency even in complex scenarios such as infrared, and occlusion. Neighbor Feature Centralization: it explores each sample's potential positive samples from its neighborhood. Experiments demonstrate that our generative model exhibits strong generalization capabilities and maintains high identity consistency. With the Feature Centralization framework, we achieve impressive performance even with an ImageNet pre-trained model without ReID training, reaching mAP/Rank-1 of 52.81/78.92 on Market1501. Moreover, our method sets new state-of-the-art results across standard, cross-modality, and occluded ReID tasks, showcasing strong adaptability.

We present Florence-VL, a new family of multimodal large language models (MLLMs) with enriched visual representations produced by Florence-2, a generative vision foundation model. Unlike the widely used CLIP-style vision transformer trained by contrastive learning, Florence-2 can capture different levels and aspects of visual features, which are more versatile to be adapted to diverse downstream tasks. We propose a novel feature-fusion architecture and an innovative training recipe that effectively integrates Florence-2's visual features into pretrained LLMs, such as Phi 3.5 and LLama 3. In particular, we propose "depth-breath fusion (DBFusion)" to fuse the visual features extracted from different depths and under multiple prompts. Our model training is composed of end-to-end pretraining of the whole model followed by finetuning of the projection layer and the LLM, on a carefully designed recipe of diverse open-source datasets that include high-quality image captions and instruction-tuning pairs. Our quantitative analysis and visualization of Florence-VL's visual features show its advantages over popular vision encoders on vision-language alignment, where the enriched depth and breath play important roles. Florence-VL achieves significant improvements over existing state-of-the-art MLLMs across various multi-modal and vision-centric benchmarks covering general VQA, perception, hallucination, OCR, Chart, knowledge-intensive understanding, etc. To facilitate future research, our models and the complete training recipe are open-sourced.

Diffusion models, especially denoising diffusion probabilistic models (DDPMs), are prevalent tools in generative AI, making their intellectual property (IP) protection increasingly important. Most existing IP protection methods for DDPMs are invasive, e.g., model watermarking, which alter model parameters and raise concerns about performance degradation, also with requirement for extra computational resources for retraining or fine-tuning. In this paper, we propose the first non-invasive fingerprinting scheme for DDPMs, requiring no parameter changes or fine-tuning, and keeping generation quality intact. We introduce a discriminative and robust fingerprint latent space based on the well-designed "crossing route" of noisy samples that span the performance border-zone of DDPMs, with only black-box access required for the diffusion denoiser in ownership verification. Extensive experiments demonstrate that our fingerprinting approach enjoys both robustness against the often-seen attacks and distinctiveness on various DDPMs, providing an alternative for protecting DDPMs' IP rights without compromising their performance or integrity.

Modern world models require costly and time-consuming collection of large video datasets with action demonstrations by people or by environment-specific agents. To simplify training, we focus on using many virtual environments for inexpensive, automatically collected interaction data. Genie, a recent multi-environment world model, demonstrates simulation abilities of many environments with shared behavior. Unfortunately, training their model requires expensive demonstrations. Therefore, we propose a training framework merely using a random agent in virtual environments. While the model trained in this manner exhibits good controls, it is limited by the random exploration possibilities. To address this limitation, we propose AutoExplore Agent - an exploration agent that entirely relies on the uncertainty of the world model, delivering diverse data from which it can learn the best. Our agent is fully independent of environment-specific rewards and thus adapts easily to new environments. With this approach, the pretrained multi-environment model can quickly adapt to new environments achieving video fidelity and controllability improvement. In order to obtain automatically large-scale interaction datasets for pretraining, we group environments with similar behavior and controls. To this end, we annotate the behavior and controls of 974 virtual environments - a dataset that we name RetroAct. For building our model, we first create an open implementation of Genie - GenieRedux and apply enhancements and adaptations in our version GenieRedux-G. Our code and data are available at https://github.com/insait-institute/GenieRedux