PVUW 2025 Challenge Report: Advances in Pixel-level Understanding of Complex Videos in the Wild

This report provides a comprehensive overview of the 4th Pixel-level Video Understanding in the Wild (PVUW) Challenge, held in conjunction with CVPR 2025. It summarizes the challenge outcomes, participating methodologies, and future research directions. The challenge features two tracks: MOSE, which focuses on complex scene video object segmentation, and MeViS, which targets motion-guided, language-based video segmentation. Both tracks introduce new, more challenging datasets designed to better reflect real-world scenarios. Through detailed evaluation and analysis, the challenge offers valuable insights into the current state-of-the-art and emerging trends in complex video segmentation. More information can be found on the workshop website: https://pvuw.github.io/.

Modeling Multiple Normal Action Representations for Error Detection in Procedural Tasks

Error detection in procedural activities is essential for consistent and correct outcomes in AR-assisted and robotic systems. Existing methods often focus on temporal ordering errors or rely on static prototypes to represent normal actions. However, these approaches typically overlook the common scenario where multiple, distinct actions are valid following a given sequence of executed actions. This leads to two issues: (1) the model cannot effectively detect errors using static prototypes when the inference environment or action execution distribution differs from training; and (2) the model may also use the wrong prototypes to detect errors if the ongoing action label is not the same as the predicted one. To address this problem, we propose an Adaptive Multiple Normal Action Representation (AMNAR) framework. AMNAR predicts all valid next actions and reconstructs their corresponding normal action representations, which are compared against the ongoing action to detect errors. Extensive experiments demonstrate that AMNAR achieves state-of-the-art performance, highlighting the effectiveness of AMNAR and the importance of modeling multiple valid next actions in error detection. The code is available at https://github.com/iSEE-Laboratory/AMNAR.

Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks

In this work, we present DEcoupLEd Distillation To Erase (DELETE), a general and strong unlearning method for any class-centric tasks. To derive this, we first propose a theoretical framework to analyze the general form of unlearning loss and decompose it into forgetting and retention terms. Through the theoretical framework, we point out that a class of previous methods could be mainly formulated as a loss that implicitly optimizes the forgetting term while lacking supervision for the retention term, disturbing the distribution of pre-trained model and struggling to adequately preserve knowledge of the remaining classes. To address it, we refine the retention term using "dark knowledge" and propose a mask distillation unlearning method. By applying a mask to separate forgetting logits from retention logits, our approach optimizes both the forgetting and refined retention components simultaneously, retaining knowledge of the remaining classes while ensuring thorough forgetting of the target class. Without access to the remaining data or intervention (i.e., used in some works), we achieve state-of-the-art performance across various benchmarks. What's more, DELETE is a general solution that can be applied to various downstream tasks, including face recognition, backdoor defense, and semantic segmentation with great performance.

ReferDINO-Plus: 2nd Solution for 4th PVUW MeViS Challenge at CVPR 2025

Referring Video Object Segmentation (RVOS) aims to segment target objects throughout a video based on a text description. This task has attracted increasing attention in the field of computer vision due to its promising applications in video editing and human-agent interaction. Recently, ReferDINO has demonstrated promising performance in this task by adapting object-level vision-language knowledge from pretrained foundational image models. In this report, we further enhance its capabilities by incorporating the advantages of SAM2 in mask quality and object consistency. In addition, to effectively balance performance between single-object and multi-object scenarios, we introduce a conditional mask fusion strategy that adaptively fuses the masks from ReferDINO and SAM2. Our solution, termed ReferDINO-Plus, achieves 60.43 \(\mathcal{J}\&\mathcal{F}\) on MeViS test set, securing 2nd place in the MeViS PVUW challenge at CVPR 2025. The code is available at: https://github.com/iSEE-Laboratory/ReferDINO-Plus.

Efficient Explicit Joint-level Interaction Modeling with Mamba for Text-guided HOI Generation

We propose a novel approach for generating text-guided human-object interactions (HOIs) that achieves explicit joint-level interaction modeling in a computationally efficient manner. Previous methods represent the entire human body as a single token, making it difficult to capture fine-grained joint-level interactions and resulting in unrealistic HOIs. However, treating each individual joint as a token would yield over twenty times more tokens, increasing computational overhead. To address these challenges, we introduce an Efficient Explicit Joint-level Interaction Model (EJIM). EJIM features a Dual-branch HOI Mamba that separately and efficiently models spatiotemporal HOI information, as well as a Dual-branch Condition Injector for integrating text semantics and object geometry into human and object motions. Furthermore, we design a Dynamic Interaction Block and a progressive masking mechanism to iteratively filter out irrelevant joints, ensuring accurate and nuanced interaction modeling. Extensive quantitative and qualitative evaluations on public datasets demonstrate that EJIM surpasses previous works by a large margin while using only 5\% of the inference time. Code is available \href{https://github.com/Huanggh531/EJIM}{here}.

VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulness

Video generation has advanced significantly, evolving from producing unrealistic outputs to generating videos that appear visually convincing and temporally coherent. To evaluate these video generative models, benchmarks such as VBench have been developed to assess their faithfulness, measuring factors like per-frame aesthetics, temporal consistency, and basic prompt adherence. However, these aspects mainly represent superficial faithfulness, which focus on whether the video appears visually convincing rather than whether it adheres to real-world principles. While recent models perform increasingly well on these metrics, they still struggle to generate videos that are not just visually plausible but fundamentally realistic. To achieve real "world models" through video generation, the next frontier lies in intrinsic faithfulness to ensure that generated videos adhere to physical laws, commonsense reasoning, anatomical correctness, and compositional integrity. Achieving this level of realism is essential for applications such as AI-assisted filmmaking and simulated world modeling. To bridge this gap, we introduce VBench-2.0, a next-generation benchmark designed to automatically evaluate video generative models for their intrinsic faithfulness. VBench-2.0 assesses five key dimensions: Human Fidelity, Controllability, Creativity, Physics, and Commonsense, each further broken down into fine-grained capabilities. Tailored for individual dimensions, our evaluation framework integrates generalists such as state-of-the-art VLMs and LLMs, and specialists, including anomaly detection methods proposed for video generation. We conduct extensive annotations to ensure alignment with human judgment. By pushing beyond superficial faithfulness toward intrinsic faithfulness, VBench-2.0 aims to set a new standard for the next generation of video generative models in pursuit of intrinsic faithfulness.

RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images

This paper presents RoGSplat, a novel approach for synthesizing high-fidelity novel views of unseen human from sparse multi-view images, while requiring no cumbersome per-subject optimization. Unlike previous methods that typically struggle with sparse views with few overlappings and are less effective in reconstructing complex human geometry, the proposed method enables robust reconstruction in such challenging conditions. Our key idea is to lift SMPL vertices to dense and reliable 3D prior points representing accurate human body geometry, and then regress human Gaussian parameters based on the points. To account for possible misalignment between SMPL model and images, we propose to predict image-aligned 3D prior points by leveraging both pixel-level features and voxel-level features, from which we regress the coarse Gaussians. To enhance the ability to capture high-frequency details, we further render depth maps from the coarse 3D Gaussians to help regress fine-grained pixel-wise Gaussians. Experiments on several benchmark datasets demonstrate that our method outperforms state-of-the-art methods in novel view synthesis and cross-dataset generalization. Our code is available at https://github.com/iSEE-Laboratory/RoGSplat.

ViSpeak: Visual Instruction Feedback in Streaming Videos

Recent advances in Large Multi-modal Models (LMMs) are primarily focused on offline video understanding. Instead, streaming video understanding poses great challenges to recent models due to its time-sensitive, omni-modal and interactive characteristics. In this work, we aim to extend the streaming video understanding from a new perspective and propose a novel task named Visual Instruction Feedback in which models should be aware of visual contents and learn to extract instructions from them. For example, when users wave their hands to agents, agents should recognize the gesture and start conversations with welcome information. Thus, following instructions in visual modality greatly enhances user-agent interactions. To facilitate research, we define seven key subtasks highly relevant to visual modality and collect the ViSpeak-Instruct dataset for training and the ViSpeak-Bench for evaluation. Further, we propose the ViSpeak model, which is a SOTA streaming video understanding LMM with GPT-4o-level performance on various streaming video understanding benchmarks. After finetuning on our ViSpeak-Instruct dataset, ViSpeak is equipped with basic visual instruction feedback ability, serving as a solid baseline for future research.

Progressive Human Motion Generation Based on Text and Few Motion Frames

Although existing text-to-motion (T2M) methods can produce realistic human motion from text description, it is still difficult to align the generated motion with the desired postures since using text alone is insufficient for precisely describing diverse postures. To achieve more controllable generation, an intuitive way is to allow the user to input a few motion frames describing precise desired postures. Thus, we explore a new Text-Frame-to-Motion (TF2M) generation task that aims to generate motions from text and very few given frames. Intuitively, the closer a frame is to a given frame, the lower the uncertainty of this frame is when conditioned on this given frame. Hence, we propose a novel Progressive Motion Generation (PMG) method to progressively generate a motion from the frames with low uncertainty to those with high uncertainty in multiple stages. During each stage, new frames are generated by a Text-Frame Guided Generator conditioned on frame-aware semantics of the text, given frames, and frames generated in previous stages. Additionally, to alleviate the train-test gap caused by multi-stage accumulation of incorrectly generated frames during testing, we propose a Pseudo-frame Replacement Strategy for training. Experimental results show that our PMG outperforms existing T2M generation methods by a large margin with even one given frame, validating the effectiveness of our PMG. Code will be released.

ChainHOI: Joint-based Kinematic Chain Modeling for Human-Object Interaction Generation

We propose ChainHOI, a novel approach for text-driven human-object interaction (HOI) generation that explicitly models interactions at both the joint and kinetic chain levels. Unlike existing methods that implicitly model interactions using full-body poses as tokens, we argue that explicitly modeling joint-level interactions is more natural and effective for generating realistic HOIs, as it directly captures the geometric and semantic relationships between joints, rather than modeling interactions in the latent pose space. To this end, ChainHOI introduces a novel joint graph to capture potential interactions with objects, and a Generative Spatiotemporal Graph Convolution Network to explicitly model interactions at the joint level. Furthermore, we propose a Kinematics-based Interaction Module that explicitly models interactions at the kinetic chain level, ensuring more realistic and biomechanically coherent motions. Evaluations on two public datasets demonstrate that ChainHOI significantly outperforms previous methods, generating more realistic, and semantically consistent HOIs. Code is available \href{https://github.com/qinghuannn/ChainHOI}{here}.