Industrial 3D face assets creation typically reconstructs topology-consistent face meshes from multi-view images for downstream production. However, high-quality reconstruction usually requires manual processing or specific capture settings. Recently NeRF has shown great advantages in 3D reconstruction, by representing scenes as density and radiance fields and utilizing neural volume rendering for novel view synthesis. Inspired by this, we introduce a novel method which combines explicit mesh with neural volume rendering to optimize geometry of an artist-made template face mesh from multi-view images while keeping the topology unchanged. Our method derives density fields from meshes using distance fields as an intermediary and encodes radiance field in compact tri-planes. To improve convergence, several adaptions tailored for meshes are introduced to the volume rendering. Experiments demonstrate that our method achieves superior reconstruction quality compared to previous approaches, validating the feasibility of integrating mesh and neural volume rendering.

Autoregressive diffusion models have enabled high-quality video generation, yet their sequential nature inherently suffers from error accumulation. In long-horizon video synthesis, minor prediction deviations compound over time, inevitably leading to unconstrained generative drift, structural collapse, and severe visual degradation. To address this, we propose Cycle-World, a novel framework designed for stable and temporally consistent long-video generation. Our approach tackles error drift by enforcing strict temporal reversibility across both the training and inference phases. Theoretically, we demonstrate that forward generative drift can be strictly bottlenecked by a cycle-consistency objective. During training, we integrate an efficient reverse-prediction model to implicitly embed causal constraints into the forward generator, compelling it to produce reversible sequences that tightly adhere to the natural video manifold. At inference time, we repurpose this frozen reverse model as a runtime corrector. Through gradient-based cycle guidance, it iteratively refines the generated latent representations, actively suppressing accumulated errors before they are committed to the historical context. Extensive experiments on the VBench benchmark demonstrate that Cycle-World's dual-phase synergy significantly mitigates error drift, achieving state-of-the-art overall generation quality and long-horizon temporal consistency in 60-second synthesis.

Contrastive Language-Image Pre-training (CLIP) has achieved widely applications in various computer vision tasks, e.g., text-to-image generation, Image-Text retrieval and Image captioning. However, CLIP suffers from high memory and computation cost, which prohibits its usage to the resource-limited application scenarios. Existing CLIP compression methods typically reduce the size of pre-trained CLIP weights by selecting their subset as weight inheritance for further retraining via mask optimization or important weight measurement. However, these select-based weight inheritance often compromises the feature presentation ability, especially on the extreme compression. In this paper, we propose a novel mapping-based CLIP compression framework, CLIP-Map. It leverages learnable matrices to map and combine pretrained weights by Full-Mapping with Kronecker Factorization, aiming to preserve as much information from the original weights as possible. To mitigate the optimization challenges introduced by the learnable mapping, we propose Diagonal Inheritance Initialization to reduce the distribution shifting problem for efficient and effective mapping learning. Extensive experimental results demonstrate that the proposed CLIP-Map outperforms select-based frameworks across various compression ratios, with particularly significant gains observed under high compression settings.

We introduce EgoSim, a closed-loop egocentric world simulator that generates spatially consistent interaction videos and persistently updates the underlying 3D scene state for continuous simulation. Existing egocentric simulators either lack explicit 3D grounding, causing structural drift under viewpoint changes, or treat the scene as static, failing to update world states across multi-stage interactions. EgoSim addresses both limitations by modeling 3D scenes as updatable world states. We generate embodiment interactions via a Geometry-action-aware Observation Simulation model, with spatial consistency from an Interaction-aware State Updating module. To overcome the critical data bottleneck posed by the difficulty in acquiring densely aligned scene-interaction training pairs, we design a scalable pipeline that extracts static point clouds, camera trajectories, and embodiment actions from in-the-wild large-scale monocular egocentric videos. We further introduce EgoCap, a capture system that enables low-cost real-world data collection with uncalibrated smartphones. Extensive experiments demonstrate that EgoSim significantly outperforms existing methods in terms of visual quality, spatial consistency, and generalization to complex scenes and in-the-wild dexterous interactions, while supporting cross-embodiment transfer to robotic manipulation. Codes and datasets will be open soon. The project page is at egosimulator.github.io.

The increasingly realistic and diverse generative data has led some deepfake detection methods to shift towards learning robust real content, \textit{e.g.}, via reconstruction-based tasks. However, most existing approaches rely primarily on prevalent continuous modeling (\textit{e.g.}, GMMs, VAEs, Diffusion Models) to construct a continuous latent manifold of real data, with the aim of improving the generalization capability, while overlooking a critical issue, \textit{i.e.}, such continuity may facilitate the interpolation of forgery artifacts, consequently causing ambiguity in detection. To alleviate this problem, we integrate discrete modeling into the feature space of the CLIP vision encoder, striking a balance between continuous manifold modeling and discrete representation. By incorporating a learnable vector quantized codebook, the real latent manifold is discretized, imposing a more stringent information bottleneck that reduces the likelihood of embedding generative artifacts. In order to further enhance the generalization of discrete modeling, we propose an adaptive tangent space projection mechanism that yields a continuous relaxation of the discrete real distribution within a controllable range. With these components, our method constructs a real distribution that is both tightly constrained and broadly generalizable, enhancing robustness to unseen forgeries. Extensive experiments on diverse datasets demonstrate the effectiveness of our method.

Identity-preserving video generation (IPVG) aims to synthesize high-fidelity videos that follow text prompts while faithfully preserving a reference identity. Despite recent progress, existing IPVG methods still struggle to balance high-level semantic control and low-level identity fidelity. To bridge this gap, we propose ST-DRC, an effective Spatial-Temporal Decoupled Reference Conditioning framework for identity-preserving text-to-video generation. At the framework level, ST-DRC performs latent in-context feature injection by encoding the reference image with the video VAE and concatenating it with noisy video latents, enabling rich low-level identity details to be accessed without additional adapters. To separate identity-aware reference retrieval from appearance copying, we introduce TASS-RoPE, a Temporal-Adjacent Spatial-Shifted RoPE scheme that places reference tokens near the video sequence in time but shifts them in space, allowing reference information to flow through spatio-temporal attention while suppressing pixel-level copy-paste shortcuts. To further prevent shortcut learning and strengthen the otherwise diluted identity supervision in the diffusion objective, we combine appearance-invariant reference augmentation with face-guided identity objectives, encouraging the model to preserve identity under variations in color, pose, and layout. At inference time, we introduce a three-stream reference classifier-free guidance strategy that independently controls text adherence and reference fidelity. Experiments demonstrate that ST-DRC achieves strong identity preservation, prompt alignment, temporal consistency, and video quality with a lightweight design built on LTX-2.3. Our method ranks among the top submissions in the facial identity-preserving video generation track, validating the effectiveness of spatial-temporal decoupled reference conditioning.

Single-image super-resolution (SISR) plays an important role in a wide range of computer vision applications. Although recent SISR methods have achieved good results, there is no effective solution to the problem of high-frequency feature loss after super-resolution due to attentional drift caused by deep attention architectures. To address these problems, we propose a novel Transformer-based network called RFASR, which pioneers the design of a global attention boosting module (PixBoost) and a Self-Graph Attention (SGA) upsampling module. The former enhances feature representation by combining meta-learning and multi-spatial channel aggregation, while the latter improves the modeling of high-frequency details through graph-based attention in the up-sampling phase. Experimental results on multiple datasets show that RFASR has a very high reconstruction level while maintaining optimal recovery efficiency.

Curvature angle is a quantitative measurement of a curve, in which Cobb angle is customized for spinal curvature. Automatic Cobb angle measurement from X-ray images is crucial for scoliosis screening and diagnosis. However, most existing regression-based and segmentation-based methods struggle with inaccurate spine representations or mask connectivity and fragmentation issues. Besides, landmark-based methods suffer from insufficient training data and annotations. To address these challenges, we propose a novel curvature angle estimation framework named CurvNet including latent contour representation based contour detection and iterative data engine based image self-generation. Specifically, we propose a parameterized spine contour representation in latent space, which enables eigen-spine decomposition and spine contour reconstruction. Latent contour coefficient regression is combined with anchor box classification to solve inaccurate predictions and mask connectivity issues. Moreover, we develop a data engine with image self-generation, automatic annotation, and automatic selection in an iterative manner. By our data engine, we generate a clean dataset named Spinal-AI2024 without privacy leaks, which is the largest released scoliosis X-ray dataset to our knowledge. Extensive experiments on public AASCE2019, our private Spinal2023, and our generated Spinal-AI2024 datasets demonstrate that our method achieves state-of-the-art Cobb angle estimation performance. Our code and Spinal-AI2024 dataset are available at https://github.com/Ernestchenchen/CurvNet and https://github.com/Ernestchenchen/Spinal-AI2024, respectively.

The rapid advancement of Artificial Intelligence Generated Content (AIGC) has revolutionized video generation, enabling systems ranging from proprietary pioneers like OpenAI's Sora, Google's Veo3, and Bytedance's Seedance to powerful open-source contenders like Wan and HunyuanVideo to synthesize temporally coherent and semantically rich videos. These advancements pave the way for building "world models" that simulate real-world dynamics, with applications spanning entertainment, education, and virtual reality. However, existing reviews on video generation often focus on narrow technical fields, e.g., Generative Adversarial Networks (GAN) and diffusion models, or specific tasks (e. g., video editing), lacking a comprehensive perspective on the field's evolution, especially regarding Auto-Regressive (AR) models and integration of multimodal information. To address these gaps, this survey firstly provides a systematic review of the development of video generation technology, tracing its evolution from early GANs to dominant diffusion models, and further to emerging AR-based and multimodal techniques. We conduct an in-depth analysis of the foundational principles, key advancements, and comparative strengths/limitations. Then, we explore emerging trends in multimodal video generation, emphasizing the integration of diverse data types to enhance contextual awareness. Finally, by bridging historical developments and contemporary innovations, this survey offers insights to guide future research in video generation and its applications, including virtual/augmented reality, personalized education, autonomous driving simulations, digital entertainment, and advanced world models, in this rapidly evolving field. For more details, please refer to the project at https://github.com/sjtuplayer/Awesome-Video-Foundations.

Existing offline feed-forward methods for joint scene understanding and reconstruction on long image streams often repeatedly perform global computation over an ever-growing set of past observations, causing runtime and GPU memory to increase rapidly with sequence length and limiting scalability. We propose Streaming Semantic Gaussian Splatting (S2GS), a strictly causal, incremental 3D Gaussian semantic field framework: it does not leverage future frames and continuously updates scene geometry, appearance, and instance-level semantics without reprocessing historical frames, enabling scalable online joint reconstruction and understanding. S2GS adopts a geometry–semantic decoupled dual-backbone design: the geometry branch performs causal modeling to drive incremental Gaussian updates, while the semantic branch leverages a 2D foundation vision model and a query-driven decoder to predict segmentation masks and identity embeddings, further stabilized by query-level contrastive alignment and lightweight online association with an instance memory. Experiments show that S2GS matches or outperforms strong offline baselines on joint reconstruction-and-understanding benchmarks, while significantly improving long-horizon scalability: it processes 1,000+ frames with much slower growth in runtime and GPU memory, whereas offline global-processing baselines typically run out of memory at around 80 frames under the same setting.

Generating intermediate video content of varying lengths based on given first and last frames, along with text prompt information, offers significant research and application potential. However, traditional frame interpolation tasks primarily focus on scenarios with a small number of frames, no text control, and minimal differences between the first and last frames. Recent community developers have utilized large video models represented by Wan to endow frame-to-frame capabilities. However, these models can only generate a fixed number of frames and often fail to produce satisfactory results for certain frame lengths, while this setting lacks a clear official definition and a well-established benchmark. In this paper, we first propose a new practical Semantic Frame Interpolation (SFI) task from the perspective of academic definition, which covers the above two settings and supports inference at multiple frame rates. To achieve this goal, we propose a novel SemFi model building upon Wan2.1, which incorporates a Mixture-of-LoRA module to ensure the generation of high-consistency content that aligns with control conditions across various frame length limitations. Furthermore, we propose SFI-300K, the first general-purpose dataset and benchmark specifically designed for SFI. To support this, we collect and process data from the perspective of SFI, carefully designing evaluation metrics and methods to evaluate the performance of the model in multiple dimensions, including images and videos, and various aspects, including consistency and diversity. Through extensive experiments on SFI-300K, we demonstrate that our method is particularly well-suited to meet the requirements of the SFI task.

Equipping Large Language Models (LLMs) with persistent memory enhances interaction continuity and personalization but introduces new safety risks. Specifically, contaminated or biased memory accumulation can trigger abnormal agent behaviors. Existing evaluation methods have not yet established a standardized framework for measuring memory misevolution. This phenomenon refers to the gradual behavioral drift resulting from repeated exposure to misleading information. To address this gap, we introduce MemEvoBench, the first benchmark evaluating long-horizon memory safety in LLM agents against adversarial memory injection, noisy tool outputs, and biased feedback. The framework consists of QA-style tasks across 7 domains and 36 risk types, complemented by workflow-style tasks adapted from 20 Agent-SafetyBench environments with noisy tool returns. Both settings employ mixed benign and misleading memory pools within multi-round interactions to simulate memory evolution. Experiments on representative models reveal substantial safety degradation under biased memory updates. Our analysis suggests that memory evolution is a significant contributor to these failures. Furthermore, static prompt-based defenses prove insufficient, underscoring the urgency of securing memory evolution in LLM agents.

Text-based human motion editing aims to modify existing motion sequences according to natural language instructions while maintaining the consistency of the original motion. Existing diffusion-based approaches often rely on heuristic similarity cues or coarse global conditioning, leading to motion distortion and suboptimal semantic alignment. The key challenge lies in balancing change (i.e. precisely editing target regions) and invariance (i.e. preserving unedited parts). To handle such challenge, we propose an Omni-Supervised Positive-Negative Learning framework, named OmniME. Our method integrates three complementary components: (1) retrospective feature supervision that enforces coarse-to-fine consistency across transformer layers,(2) motion preservation mechanism that focuses on subtle variations accoding to the source-target similarity, and (3) triplet-based semantic alignment that strengthens text-motion correspondence. Together, these components form a unified supervision paradigm that balances change and invariance. Extensive experiments on the MotionFix and STANCE Adjustment datasets demonstrate that OmniME achieves state-of-the-art performance in editing alignment , validating the effectiveness of our unified learning framework. The code will be made publicly available uppon acceptance.

Night-time scene parsing is important for practical applications such as autonomous driving and robot vision. Since annotating is time-consuming, Unsupervised Domain Adaptation (UDA) is an effective solution for night-time scene parsing. Due to the low illumination, over/under-exposure, and motion blur in night-time scenes, existing methods can not connect daytime scenes and night-time scenes well, limiting their performance. Some methods rely on day-night paired images, which are costly to collect and therefore impractical. In this paper, we propose DANIM, a self-training UDA network for night-time scene parsing. We introduce an intermediate domain that explicitly models the connection between daytime scenes and night-time scenes from lighting and structure. The intermediate domain shares similar structure information with the night-time target domain and similar lighting information with the daytime source domain. By harnessing the rich prior knowledge of a pre-trained text-driven generative model, the intermediate domain can be generated and we propose a scoring mechanism for selecting the high-quality one for training. Besides, we propose intermediate domain masking to address the inconsistency between the intermediate domain and the target domain. We further design a coupled mask strategy to make the mask more effective. Extensive experiments show that DANIM has achieved first place on the DarkZurich leaderboard and outperforms state-of-the-art methods on other widely used night-time scene parsing benchmarks, i.e. ACDC-night, NightCity, and NighttimeDriving.

Dynamic driving scene reconstruction is of great importance in fields like digital twin system and autonomous driving simulation. However, unacceptable degradation occurs when the view deviates from the input trajectory, leading to corrupted background and vehicle models. To improve reconstruction quality on novel trajectory, existing methods are subject to various limitations including inconsistency, deformation, and time consumption. This paper proposes LidarPainter, a one-step diffusion model that recovers consistent driving views from sparse LiDAR condition and artifact-corrupted renderings in real-time, enabling high-fidelity lane shifts in driving scene reconstruction. Extensive experiments show that LidarPainter outperforms state-of-the-art methods in speed, quality and resource efficiency, specifically 7 x faster than StreetCrafter with only one fifth of GPU memory required. LidarPainter also supports stylized generation using text prompts such as "foggy" and "night", allowing for a diverse expansion of the existing asset library.

This survey reviews feed-forward, point-cloud 3D reconstruction methods from DUSt3R to VGGT and their recent variants. Here, feed-forward primarily refers to predicting dense geometry and, when applicable, camera poses through learned network inference, without relying on classical per-scene SfM + MVS optimization as the main inference mechanism. We first formalize the reconstruction task in pose-aware and pose-free settings, and contrast feed-forward point-map regression with classical Structure-from-Motion and Multi-View Stereo pipelines. Building on this, we organize existing methods into three stages: early pairwise models typified by DUSt3R, DUSt3R-style extensions that enhance multi-view consistency, streaming, efficiency, and dynamic-scene handling, and large unified transformers such as VGGT that process tens to hundreds of views jointly, while noting differences in their inference paradigms. We analyze these models along shared axes, including scene representation, correspondence reasoning, pose regression, fusion strategies, and the role of large-scale training data. We summarize widely used 3D datasets and evaluation metrics, and provide a case study on the DTU benchmark for multi-view depth and point map estimation, highlighting accuracy-efficiency trade-offs between optimization-based and feed-forward approaches. Finally, we discuss open challenges in data scarcity, sparse-view reconstruction, non-Lambertian structures, dynamic scenes, long-context processing, and resource-efficient deployment, and outline future directions that combine feed-forward architectures with differentiable rendering, generative priors, and safety mechanisms to enable scalable and trustworthy 3D reconstruction systems.

The body movements accompanying speech aid speakers in expressing their ideas. Co-speech motion generation is one of the important approaches for synthesizing realistic avatars. Due to the intricate correspondence between speech and motion, generating realistic and diverse motion is a challenging task. In this paper, we propose MMoFusion, a Multi-modal co-speech Motion generation framework based on the diffusion model to ensure both the authenticity and diversity of generated motion. We propose a progressive fusion strategy to enhance the interaction of inter-modal and intra-modal, efficiently integrating multi-modal information. Specifically, we employ a masked style matrix based on emotion and identity information to control the generation of different motion styles. Temporal modeling of speech and motion is partitioned into style-guided specific feature encoding and shared feature encoding, aiming to learn both inter-modal and intra-modal features. Besides, we propose a geometric loss to enforce the joints' velocity and acceleration coherence among frames. Our framework generates vivid, diverse, and style-controllable motion of arbitrary length through inputting speech and editing identity and emotion. Extensive experiments demonstrate that our method outperforms current co-speech motion generation methods including upper body and challenging full body.

Despite progress in instruction-based video editing, unimodal textual instructions inherently struggle to convey fine-grained textures and complex dynamics. To bridge this perceptual gap, we propose Visual In-context Editing, a new paradigm elevating video editing from textual instructions to multi-modal visual guidance encompassing single image, image pair, and video pair. To facilitate this paradigm, we curate VicEdit-400K, the first large-scale dataset for visual in-context video editing. We develop an automated pipeline to generate 400K high-quality samples across ten task types, ensuring superior visual fidelity and semantic consistency through multi-dimensional filtering. Leveraging this foundation, we introduce VicEdit, a unified framework to bridge visual and textual contexts. To adaptively extract editing semantics from heterogeneous references, we design Modality-Adaptive Semantic Distillation, which produces modality-specific semantic tokens from visual references. These tokens are then synergistically integrated with textual instructions through Dual-Context Injection, enabling the generation process to benefit from both visual and textual signals. Extensive evaluations on VicEditBench demonstrate that VicEdit achieves state-of-the-art performance across both basic instruction editing and visual in-context editing tasks, establishing visual in-context learning as a powerful and controllable paradigm for video editing.

Mirror detection is crucial for avoiding collisions and misrecognition of reflections in real-world scenes. Existing methods often rely on assumptions, such as semantic similarity between objects and their reflections or contextual correlations, which may not hold in cluttered environments, limiting generalization. Recent advances in foundation models like the Segment Anything Model (SAM) have boosted general-purpose segmentation, yet SAM struggles with mirrors due to prompt ambiguity and reflective complexity. Moreover, Prior methods neither exploit SAM nor generalize well to diverse scenes. To address this, we adapt SAM for mirror segmentation by introducing a Frequency-Chirality Adapter (FCA), which encodes frequency-domain textures and visual chirality to enhance mirror-specific perception. Additionally, we design a Prior-Aware Localization (PAL) module to provide automatic prompt guidance from effective semantic priors, eliminating the need for manual inputs. Experiments on PMD, MSD, and RGBD-Mirror datasets show that our method consistently outperforms previous state-of-the-art approaches, significantly enhancing SAM’s performance on mirror segmentation.

The domain shifts often hinder accurate medical image segmentation, particularly when models trained on a source dataset are applied to a heterogeneous target dataset. To address this challenge, we propose a Dual-Path Attention Network (D ^2 N) that explicitly incorporates domain adaptation. The framework integrates a Gradient Reversal Mechanism (GRM) to encourage domain-invariant feature learning, along with a Cross-Domain Attention Module (CDAM) that enhances the fusion of global contextual information and local boundary details, thereby improving segmentation accuracy. To comprehensively evaluate the proposed model, we conducted experiments across four medical imaging datasets spanning diverse clinical applications: skin cancer segmentation (ISIC2018 and PH2), polyp segmentation and brain tumor segmentation. D ^2 N achieved strong performance, with Dice scores of 94.0% on ISIC2018, 92.1% on PH2, 96.55% on the polyp dataset and 91.5% on the brain tumor dataset. When compared against existing state-of-the-art methods, D ^2 N consistently achieved better performance than competing approaches, with superior Dice, IoU and other metric scores across all datasets. These findings demonstrate that D ^2 N effectively mitigates the impact of domain shift, substantially improves segmentation precision and generalizes well across a range of imaging modalities and anatomical structures. The code is available in the GitHub repository at https://github.com/nooriahmed/D2N-Dual-Path-Attention .