跳到正文
10月1日周四
  2. arXiv · 三维交互与编辑

    AssemblyWorld: Rethinking 3D Assembly with General-Purpose Agents

    The task of 3D assembly requires translating an understanding of parts and their relationships into precise spatial arrangements. Can pretrained general-purpose agents assemble objects through visual interaction without additional assembly-specific fine-tuning? To investigate this question, we introduce AssemblyWorld, an interactive 3D environment in which agents inspect rendered views and manipulate supplied rigid parts, guided by images or assembly manuals when available. Agents perceive part geometry through 2D views rather than direct access to mesh vertices or faces, while their resulting assemblies are evaluated geometrically. Building on this environment, we construct AssemblyWorldBench, comprising 100 assembly tasks across 80 objects spanning furniture, industrial assembly, and fracture reassembly. Evaluating eight agent systems reveals substantial differences in their capabilities. The strongest system achieves 80.9% part accuracy but 59.4% complete-assembly success. The evaluated open-source systems lag substantially behind their stronger closed-source peers in both execution reliability and assembly accuracy. Analyses of visual references, interaction trajectories, and failures show how agents revise assemblies while leaving residual positioning errors. AssemblyWorld provides a common setting for both assessing the capabilities of interactive assembly agents and characterizing the gap between approximate structure recovery and precise reconstruction.

  3. Microsoft Research

    Forecasting space weather risks on power grids

    At a glance End-to-end forecasting: A machine learning pipeline uses forecast-time solar-wind information to generate location-specific risk estimates for 66,935 substations in the continental United States. Physics and place: The system combines Auroral Electrojet (AE) and Disturbance Storm Time (Dst) forecasts with local latitude, geology, and ground conductivity. Advance warning: The pipeline detected nearly 80% of major space-weather events during the evaluation period and can warn grid operators 30 to 60 minutes before a specific risk appears. Forecasting a threat to critical infrastructure During the May 2024 geomagnetic storm, utilities across North America prepared for possible impacts as auroras extended far beyond their usual range. The storm would degrade GPS accuracy and satellite operations, impacting farm operations in North America and Europe. Modern society depends on reliable electric power, yet extreme space-weather events can induce currents in transmission networks that damage equipment and increase operational risk. The challenge is not only knowing that a storm is approaching, but estimating when and where its effects could be most severe with enough warning for grid operators to respond. During my summer internship at Microsoft Research, I developed a machine learning system that forecasts space-weather risk across 66,935 substations in the continental United States. The system measures how the sun’s activity can affect the earth’s magnetic field and, ultimately, the power grid. It combines solar-wind observations, forecasts of the Auroral Electrojet (AE) and Disturbance Storm Time (Dst) indices, physics-informed constraints, local geological conductivity, and grid-infrastructure data to produce location-specific risk estimates 30-60 minutes ahead of potential impact. Building an end-to-end prediction pipeline Space-weather prediction is difficult because it spans several coupled systems. The solar wind changes rapidly, its interaction with Earth’s magnetosphere is irregular, and the resulting ground effects depend on local conditions. Regions with resistive bedrock can experience stronger geomagnetically induced currents (GICs) than regions with more conductive geology. Transmission-line orientation, latitude, and other power-system characteristics further influence the exposure of individual assets. The pipeline addresses this complexity in three stages, shown in Figure 1. First, solar-wind measurements from the L1 Lagrange point are used to generate forecasts of the AE and Dst indices, while geological conductivity and location features are assembled for each substation. Second, a gradient-boosting model combines these forecasted and location-specific inputs to estimate dB/dt, the rate of magnetic-field change associated with GIC risk. Finally, the resulting predictions are converted into location-specific risk estimates and aggregated into a continental risk assessment. A system of 50 AI agents helped explore features, validation strategies, and model configurations across the pipeline. Only public data sources were used, included NASA OMNI and NASA-aggregated Kyoto World Data Center data, INTERMAGNET and U.S. Geological Survey magnetometer observations, and GridSFM-derived grid data. Figure 1. End-to-end forecasting pipeline from L1 solar-wind measurements to location-specific risk estimates across 66,935 U.S. substations. Evaluating geomagnetic forecasts and infrastructure risk The AE predictor was designed to forecast rare, high-intensity geomagnetic activity that drives infrastructure risk. In the 2020-2026 evaluation period, the model produced forecasts spanning nearly the full observed range of AE activity and outperformed several empirical solar-wind-based approaches. The Dst predictor provided an additional signal describing large-scale geomagnetic storm strength. During the most geomagnetically active periods of the 2020-2026 evaluation period, the machine-learning model outperformed the Burton equation on 62.2% of individual hours. The model also produced a substantially wider prediction range than Burton-style approaches and improved severe-event detection in the end-to-end forecasting system by 1.2 percentage points when combined with AE forecasts. Predictor Result Baseline comparison Impact AE 410.2 nT RMSE Lower RMSE than empirical and solar-wind-only baselines Captures extreme-event activity used by the GIC forecasting system Dst 7.2 nT RMSE Outperformed Burton on 62.2% of peak-activity hours and provided wider prediction range +1.2pp severe storm detection when integrated into final GIC system The GIC risk stage was evaluated differently. There is no equivalent widely deployed operational system that provides a direct industry benchmark for this calculation, so the machine learning model was compared with simple linear regression. As Figure 2 shows, the system achieved detection rates of 76.5% for major events (≥10 nT/min), 81.2% for severe events (≥20 nT/min), and 64.1% for extreme events (≥50 nT/min). False-alarm rates increased with storm severity, reflecting the trade-off between missed events and cautious alerts. Performance varied by latitude, with the highest detection rates at northern stations where geomagnetic activity is strongest. Figure 2. Detection rates across storm-severity thresholds for regionally different substations, with a comparison between the average POD for each severity tier. Translating predictions into continental risk assessments The final stage translates predicted geomagnetic activity into location-specific estimates of dB/dt, the rate of magnetic-field change associated with GIC exposure. Rather than issuing a single alert for the continental United States, the system combines storm conditions with each substation’s latitude and geological factor. This produces continuous risk estimates that can distinguish lower-risk locations from areas where resistive geology can amplify ground-level effects. Figure 3 illustrates the output for a representative major-storm scenario. The map is a demonstration of the model’s continental-scale output, not a record of a live operational event. It shows how a grid operator or planner could move from a broad space-weather warning toward a more targeted view of which locations may warrant closer analysis. The current pipeline produced estimates for all 66,935 substations in approximately 333 milliseconds during measured inference, allowing many scenarios to be evaluated quickly. Figure 3. Demonstration of a continental GIC risk assessment under a representative major-storm scenario. Colors indicate modeled risk levels across 66,935 substations. Implications and looking forward This work demonstrates how physics-grounded machine learning could support more specific and timely assessment of space-weather exposure. Earlier, location-specific information could help utilities prioritize engineering review and consider targeted protective actions, such as adjusting reactive-power reserves or temporarily reconfiguring parts of the network. Further validation with utilities and operational data would be needed before the system could be used in grid operations. The project also connects with broader Microsoft Research work on AI for power systems. The open grid-data pipeline provided realistic U.S. transmission models, while GridSFM applies deep learning to AC optimal power flow for fast scenario analysis. Together, these efforts point toward richer planning and resilience workflows that combine hazard forecasts, grid topology, and power-flow analysis. Extend forecast horizons: Explore temporal-transformer approaches that capture longer-range patterns in solar-wind data and move beyond the current 30-60-minute window. Scale internationally: Adapt the system to additional regions while accounting for different geological conditions and grid topologies. Integrate with grid operations: Evaluate how forecasts could support existing decision-making and engineering-review workflows before considering higher levels of automation. Provide transformer-level risk: Add asset-specific characteristics to move from substation-level estimates toward more granular assessments of critical equipment. Related Microsoft Research work GridSFM: A new, small foundation model for the electric grid Building realistic electric transmission grid dataset at scale: a pipeline from open dataset Acknowledgements Special thanks to my mentors Weiwei Yang and Spencer Fowers for their guidance throughout this project, and to Amber Hoak , Weishung Liu , and Andrea Britto for their editorial and technical feedback. Opens in a new tab The post Forecasting space weather risks on power grids appeared first on Microsoft Research .

9月30日周三
  1. arXiv · 角色动作与重定向

    ECHO-G: Embodied Co-speech Humanoid mOtion Generation

    Generating full-body co-speech motion for humanoid robots requires coordinating speech prosody, linguistic content, and embodiment-specific motion. To this end, we present ECHO-G, a framework jointly conditioned on speech audio and timed transcripts. Its Speech-Grounded Diffusion Transformer (SGDiT) combines frame-aligned acoustic features with token-level linguistic context, preserving their distinct granularities. Trained with rectified flow matching, it models one-to-many utterance-motion relationships directly in robot space. To support training and evaluation, we introduce a BEAT2-derived audio-text-robot dataset and a benchmark covering co-speech characteristics, robot-motion quality, and runtime efficiency. Comparative evaluation supports direct robot-space generation over the evaluated human-motion generation and retargeting pipelines, while modality ablations highlight the benefits of joint audio-text conditioning. We further demonstrate deployment on a physical humanoid robot. A complementary video-rating study also favors joint conditioning over the alternatives. The dataset and training, inference, and evaluation code are available through our project page.

  2. GitHub · sparkjsdev/spark

    sparkjsdev/spark 发布 v2.3.0

    2.3.0 (Sep 29, 2026) Bug fixes and improvements Enhancements Fix / Enable on-demand rendering: Render a frame only when explicitly requested ( #449 , #451 , #452 ) (fix #298 ) ( @asundqui , @onderilkesever , @mrxz , @oscarlorentzon ) Make members of SplatEncoding non-optional to avoid repeated fallback handling ( #415 ) ( @mrxz ) Tree shaking improvements by focussing on classes that users might not need or use. ( #417 ) ( @mrxz ) Update dev dependencies: TypeScript and Vite, use Vitest for tests ( #445 ) ( @mrxz ) Use instanceof instead of isTexture flag for THREE.js type checking ( #446 ) ( @oscarlorentzon ) Add typechecking to CI ( #427 ) ( @oscarlorentzon ) Take extended splats ( ExtSplats ) when writing SPZ. writeSpz only took PackedSplats ( #440 ) ( @oscarlorentzon ) Accept cookie and header options when loading splats ( #459 ) ( @oscarlorentzon ) Use the calling renderer's encoding for its splats. Fix issue when using two SparkRenderers as when rendering two different viewpoints simultaneously ( #453 ) ( @oscarlorentzon ) Run the Rust tests in Linux CI ( #444 ) ( @oscarlorentzon ) Docs improvements ( @mrxz , @oscarlorentzon , @mrxz ) Improve tests ( @mrxz , @oscarlorentzon , @mrxz ) Bug Fixes Fix a GPU resource memory leak. Dispose geometry and material in SparkRenderer.dispose() ( #410 ) ( @ArisLiWind , @mrxz ) Fix references to repo URL ( #426 ) ( @oscarlorentzon ) Write non-finite SPZ rotations as the identity. Issue when writting SPZ v3 ( #454 ) ( @oscarlorentzon ) Allocate missing SH3 data texture for paged ext splats. Addresses RangeError: offset is out of bounds error ( #424 ) ( @oscarlorentzon ) Await WebAssembly initialization and check sh degress in transcodeSpz ( #438 ) (fix #437 ) ( @mrxz ) Await for WebAssembly initialization before writing SPZ ( #448 ) ( @oscarlorentzon ) Make built-in controls rotate around the controlled object's up axis instead of world y axis ( #434 ) (fix #300 ) ( @dev-xdh , @oscarlorentzon ) Read and apply maxSh passed in PagedSplats options. Previously ignored and SplatPager limit applied instead ( #447 ) ( @oscarlorentzon ) Fix wrong raycaster direction for pinch and press movement under a rotated parent ( #435 ) ( @oscarlorentzon ) Controls: Skip pointer frames that took no time ( #436 ) ( @oscarlorentzon ) Apply default encoding when none is given when writting PackedSplats ( #439 ) ( @oscarlorentzon ) Make rotation speed of keys and gamepad (XR controllers) consistent ( #442 ) ( @oscarlorentzon ) Regression. Restore v3 as the default version when outputting SPZ format ( #441 ) ( @oscarlorentzon ) Cancel scheduled work (spark updates, splat sorting) when disposing the renderer ( #450 ) ( @oscarlorentzon ) 官方 Release：v2.3.0。请查看 Release notes、依赖和许可证变化。

  3. GitHub · ronak-create/FableCut

    ronak-create/FableCut 发布 FableCut v1.9.0 — targeting & locks, trim tools, markers & full JKL

    Added Named, coloured markers — markers now carry an optional label and color (gold · red · orange · green · cyan · blue · purple · pink) and show their name on the ruler. Drag a marker to move it (it snaps), click it to jump, double-click or right-click to rename / recolour / delete. A Markers button lists every marker to jump to or edit, with Clear all. ⇧M / Alt+⇧M go to the next / previous marker. Marker edits are now undoable. Snap targets — the ▾ beside Snap picks what pulls a drag in: clip edges, playhead, markers, IN/OUT, keyframes (new, off by default) and the frame grid (rounds to whole frames when nothing else is in reach). A cyan guide line shows where a drag snapped. Remembered per browser. Typed timecode — click the playhead readout (or just type a digit) and enter a time: 01:15:00 , packed digits ( 1500 = 15 s), +30 / -1:00 to offset, or seconds ( 12.5 ). The IN and OUT readouts take typed times too, in both monitors. Full JKL — J now plays in reverse (silent, frame by frame); tapping J or L again speeds up in that direction and the other key turns around at 1×. Hold K and tap J / L to step one frame, or hold both to crawl at ¼ speed. Works in the Source monitor too. Trim tools — a tool picker in the timeline toolbar with Premiere's shortcuts where they are free: Selection ( V ), Ripple edit ( B ), Rolling edit ( R — Premiere's N is Snap here), Slip ( Y ) and Slide ( U ). Ripple and Roll act on clip edges, Slip and Slide on the whole clip. All are clamped to the source media, MIN_DUR and free room, move linked partners together, respect targeting and locks, snap like normal trims, undo in one step, and show the offset beside the pointer while dragging. Lift ( ; ) and Extract ( ' ) — remove the IN→OUT range on the targeted tracks; Extract also closes the gap. Buttons in the timeline toolbar too. Track targeting — click a track's name to target or untarget it. Split, insert, replace, ripple delete, close gap, next gap, T / ⇧T and jump-to-cut touch targeted tracks only, and Source placement lands on the targeted lanes. A selection is still edited wherever it sits. Stored as project.untargetedTracks . Track lock — a padlock in the track header ( project.lockedTracks ). Nothing on a locked track can be moved, trimmed, split, deleted or edited in the inspector, and ripple edits leave it in place while the other tracks shift. Clip lock and clip disable — right-click a clip or use the new toggles at the top of the inspector. A locked clip ( locked: true ) is pinned like a locked track; a disabled clip ( disabled: true , ⇧E ) stays on the timeline but is left out of preview, audio and export. A linked A/V group with any locked member counts as locked as a whole, so sync never breaks. Unlink / link audio and video ( Ctrl/Cmd+L , or the clip menu). Unlinking stamps unlinked: true so the reload-time relink no longer pairs the clips back up; Link re-joins a video and its audio once they line up. Agents respect locks — fablecut_patch_project refuses updateClip / removeClip on a locked clip (or one linked to it) and addClip onto a locked track, unless the op carries force: true . set:{locked:null} always works, and setProject accepts lockedTracks / untargetedTracks . The compact project view lists locked / untargeted tracks and flags clips [locked] , [disabled] , [unlinked] . Changed K only stops playback now (it used to toggle play); Space still toggles. ⇧M goes to the next marker instead of adding one — plain M adds / removes. The track enable toggle (eye / speaker) is now output-only : a disabled track is still hidden from preview and export, but edits no longer skip it — targeting decides that. As a side effect, soloing a track no longer blocks edits on the others. The media bin's + Title / + Adjust / + Import / + URL buttons share one row; the "Assets" heading beside them is gone. The editor's browser tab reads "FableCut - Video Editor". Fixed Exports are now tagged with their colour primaries and transfer (BT.709 by default). Every profile used to write them as "unknown", so some players guessed the colours; ProRes lost its matrix tag too. The tags are stamped with setparams , which also works on FFmpeg builds older than 7.1. Full changelog: v1.8.0...v1.9.0 官方 Release：v1.9.0。请查看 Release notes、依赖和许可证变化。

  4. arXiv · 角色动作与重定向

    GestAdapt: Workspace-Conditioned Co-Speech Gesture Generation for Humanoid Robots

    Co-speech gestures for robots must adapt not only to speech and embodiment, but also to the workspace available for performing the motion. Since the same speech can be accompanied by different gestures, a robot can respond to workspace constraints, e.g., gestures for speech next to a wall. In these scenarios, the robot should gesture in a suitable motion rather than simply correcting an unconstrained one. To achieve this goal, we present GestAdapt, a workspace-conditioned framework that conditions co-speech gesture generation on a prescribed wrist workspace. The GestAdapt framework learns from six complementary co-speech corpora through a shared motion representation and supports retargeting to different robot embodiments. Quantitative evaluation shows that generated motions remain close to the real-motion distribution while respecting the workspace. In a user study, gestures generated under modified workspace constraints receive a mean quality score of 3.24/5, above our no-workspace variant (2.43/5) and below the reference motions (3.68/5). In a real robot evaluation, all compared motions are retargeted to the Reachy2 humanoid robot under identical workspace constraints. Motions generated with our framework rank first in 69.7\% of comparisons, higher than our no-workspace variant baseline and retargeted ground-truth motions constrained afterward. Overall, the results support adapting gestures to the available workspace during generation, rather than modifying unconstrained trajectories afterward to satisfy workspace constraints, potentially compromising gesture naturalness.

  5. arXiv · 角色动作与重定向

    Counterfactual Video Generation Enables Scalable Humanoid Loco-Manipulation

    Teaching humanoids loco-manipulation skills, such as carrying diverse objects, via visual imitation is a promising path toward generalist robots. However, collecting diverse, high-quality interaction videos, such as clips that clearly show a person's full body and unoccluded interactions with objects, poses a practical barrier to scaling this approach. We propose PRISM, a real-to-sim-to-real framework that overcomes this limitation by amplifying a handful of real videos into a large, diverse training set. PRISM first generates hundreds of diverse "counterfactual" human-object interaction videos via video-to-video (V2V) generation from a few exemplar real videos. Our contact-anchored real-to-sim pipeline then reconstructs both human and object motions, retargeting this imperfect video data into physically plausible trajectories. The intra-class variability across these counterfactual videos lets us train a single policy that generalizes to unseen objects within each category. We demonstrate the full pipeline by deploying this policy on a real robot without any real-world fine-tuning. Using only onboard depth observations, our humanoid picks up, carries, and drops objects, including boxes, barrels, bins, and balls, across novel instances, sizes, and initial configurations.

9月29日周二
  1. arXiv · 三维交互与编辑

    WINGS: Reference-Free Gaussian Splatting Inpainting with 3D-Native Generative Priors

    Inpainting 3D Gaussian Splatting scenes, a key challenge in 3D editing, requires generating plausible content within a masked region of 3D space. Prior approaches rely on 2D diffusion models to produce one or several inpainted reference views, making them susceptible to challenges associated with multi-view inconsistency and lengthy optimization times. Departing from these approaches, we introduce a reference-free Gaussian splatting inpainting method operating natively in 3D. Our method leverages the embedding space of a large, pre-trained 3D prior, combined with a structure completion network to feed a generative prior which reconstructs the missing region's geometry and appearance. Performing content generation entirely in 3D, it avoids the need to reconcile inconsistencies of multiple inpainted reference images, and is faster than related 2D-based methods. We demonstrate the effectiveness of our method qualitatively and quantitatively, through extensive experiments and a user study. To the best of our knowledge, this work is the first Gaussian splatting inpainting method to operate in the learned representation space of a 3D-native generative prior without relying on inpainted reference views.

  2. arXiv · 角色动作与重定向

    Generative Interactions: Weaving Multiparty Human Motion with Bilevel Latent Dynamics

    Human social behaviour is not a collection of independent motions, but a jointly organised process in which group dynamics and individual variation continuously shape one another. Yet existing social motion models often prioritise plausible trajectories while leaving interaction state implicit, limiting their ability to transfer across groups, tasks, and partial-observation regimes. To address this gap, we introduce Bilevel Representations for Agent Interaction Dynamics (BRAID), a hierarchical sequential latent-variable model for generative multi-person interaction. BRAID explicitly formulates social motion generation as a meta-transfer learning problem: shared interaction priors are learned across datasets and adapted through arbitrary context sets of observed people and joints. The model represents each scene through a group-level latent state that captures shared interaction dynamics and person-level latent states that capture individual behaviour conditioned on the evolving group context. This modelling choice enables coherent generation under full, sparse, or partial observations while exposing compact social-state vectors that can serve as an interface for downstream embodied-agent systems. We evaluate BRAID under a unified SMPL-based representation on social forecasting, tracking and in-filling, and response generation, using metrics that assess not only reconstruction accuracy but also realism, diversity, temporal alignment, and interpersonal coordination. We further analyse the hierarchical latent space, showing that it captures separable group- and individual-level structure.

  3. GitHub · ronak-create/FableCut

    ronak-create/FableCut 发布 FableCut v1.8.0 — 3-point editing, WebCodecs export, dynamic tracks

    The editing release — 21 merged pull requests . FableCut gets proper 3-point editing with a Source monitor, a hardware-encoded WebCodecs export, reframed delivery, and tracks you can add, remove, solo and pan. Highlights 3-point editing — a Source monitor with its own In / Out, Insert ( , ), Replace ( . ) and Ripple delete ( ⇧Del ), all sync-locked so linked audio and video never drift apart; ↑ / ↓ jump between cuts (thanks @PlkMarudny , #74 ) WebCodecs export — the browser hardware-encodes H.264 and the server stream-copies it into MP4, with bitrate and VBR / CBR in the Export dialog (thanks @PlkMarudny , #49 ) Faster Fast export — frames are snapshotted as ImageBitmaps and uploaded in batches instead of waiting on every request (thanks @PlkMarudny , #49 , #72 , #75 ) Encoding profiles — Draft, Delivery, High quality 4:2:2 10-bit, Broadcast 1080i50 and ProRes 422 HQ, defined in encoding-profiles.json and listed over MCP by fablecut_encode_profiles (thanks @PlkMarudny , #48 ) Export frame — cut a 9:16 reel out of a 16:9 edit with a delivery crop you drag on the monitor (thanks @PlkMarudny , #46 , #66 ) Export IN–OUT — every engine can render just the marked range (thanks @PlkMarudny , #68 ) Tracks — add, remove and solo video and audio tracks (up to 16 of each), per-clip stereo pan, and an L/R master meter (thanks @PlkMarudny , #41 ) Import from URL — a + URL button in the editor, and fablecut_import_media now accepts https:// links, behind SSRF guards (thanks @PlkMarudny , #63 ) Editable keyframes — change a keyframe's value from the inspector, Shift-click a label to remove it (thanks @PlkMarudny , #70 ) Ctrl/Cmd-click a slider to reset it to its default (thanks @PlkMarudny , #73 ) New animated SVG overlays: spawning-hearts and ai-star (thanks @PlkMarudny , #78 , #79 ) A behavioural test suite — MCP, REST API and SVG-library tests, run on Node 18 / 20 / 22 for every pull request ( #61 ) Fixed Fast and WebCodecs export no longer repeat frames, and no longer fail with "tainted canvas" on animated SVG overlays MCP initialize negotiates a protocol version the server actually supports instead of echoing unknown ones (thanks @michaelxer , #59 ) The master meter now matches the track meters and measures real program LUFS (thanks @PlkMarudny , #67 ); heart-pop.svg lost a stray opening shape ( #69 ) The audio graph is fully torn down on project reload instead of leaving nodes wired to live buses Performance The playhead moves with a CSS transform, so playback no longer causes layout shifts (thanks @PlkMarudny , #71 ); audio meters are denser and resize with the panel ( #65 , #80 ) The fablecut.mcpb bundle is back down to ~0.5 MB Heads-up for API callers POST /api/export/begin now requires fps instead of defaulting to 30. project.json gains tracks and panSchema . Projects saved before pan are migrated once when opened; older builds ignore both keys. Full notes in CHANGELOG.md . Thanks to everyone who sent a PR this cycle — @PlkMarudny and @michaelxer . Full Changelog : v1.7.0...v1.8.0 官方 Release：v1.8.0。请查看 Release notes、依赖和许可证变化。

  4. arXiv · 角色动作与重定向

    Why Cross-Skeleton Retargeting Is Non-Identifiable: Structural Limits of Generative Motion Models

    Cross-skeleton motion generation trains generative models to carry action structure and motion intention from one body to another. Yet a target motion that shows the right action has two explanations that the training data cannot tell apart: the model transferred the source clip, or it recovered a typical motion for the requested action. We show that this ambiguity is structural rather than incidental: under standard generative objectives, the source-conditioned retargeting map is non-identifiable in sparse heterogeneous motion domains. Unpaired distribution matching yields gauge non-identifiability: the latent spaces of different skeletons can be transformed relative to one another without changing the training evidence, so different source-conditioned maps fit it equally well. Sparse paired supervision admits the complementary failure mode, \emph{conditional-mean degeneration}: when clips are paired only by action, squared-error training converges to an average target motion that ignores the source clip. To make the missing evidence observable, we introduce Source-Instance Fidelity (SIF), a diagnostic that tests whether outputs differ from one another the way their source clips do, with the target skeleton and action held fixed. Under this diagnostic, methods that succeed at the standard action-level test on animal motion data often sit at the source-blind floor, while the methods that rise above it retain only a partial relational signal. Retargeting therefore needs objectives and evaluations that can identify the source-conditioned map it claims to learn. Project page: https://cross-skeleton-retargeting.netlify.app/.

  5. GitHub · deepseek-ai/deepseek-harness

    deepseek-ai/deepseek-harness 发布 v0.2.0-rc.2

    中文 | English ✨ 新增功能 macOS／Windows 桌面端可在菜单栏中管理和安装 dsh 命令，支持管理插件，无需另装 Node 或 pnpm。 @tianyicui 🐛 问题修复 修复新建终端菜单重复列出同名 shell 的问题。 @LegGasai 修复切换会话或返回对话后计划审阅无法打开、「查看全文」消失的问题。 @LegGasai 修复设置页关闭「显示代码工作视图」后无法调整 Agent 预设的限制。 @lsdsjy 修复持久 PowerShell 在完成状态后带有空格时无法正确识别命令结束、丢失退出码或泄露内部标记的问题。 @turtle2099 修复 macOS Intel 版桌面端内置 Node 的签名权限，避免 Office 技能命令启动时崩溃。 @07akioni 修复 macOS 和 Linux 从图形入口启动桌面端时缺少登录 shell 环境的问题，使工具路径、代理等用户环境配置可供会话使用。 @lsdsjy 🎨 体验优化 模型选择器在模型较多时提供搜索，支持模糊匹配和键盘选择。 @MrCroxx 侧栏文件页可直接用本地应用打开当前文件夹，并记住应用选择。 @yudshj 优化聊天耗时、过程信息、字号和深色主题样式，优化动画运行开销。 @yixiangihsiang , @imccyu 精简插件安装引导，并区分已安装、不兼容和内置插件的升级提示。 @Yifffan , @turtle2099 自动化任务投递的提醒改为明确标注的用户定时消息，不再要求 Agent 仅将其作为不可信提醒内容转述。 @Chinesezjc 加强 Bash 与 PowerShell 的工具提示，提醒 Agent 在删除或移动前核对实际目标路径，以避免路径误判。 @turtle2099 ⚠️ 其他变更 Windows 沙箱权限脚本改为经授权后一次完成诊断与修复，保留修改前备份和恢复命令。 @Elevator14B 更新第三方模型目录与兼容适配至 pi-ai 0.87.1；部分旧模型 ID 被移除，已保存的选择可能需要重新选择。 @tianyicui 实验性添加异步问答模式，需要手动配置开启：等待超时后 Agent 可继续独立工作，用户仍可稍后回答。 @Magolor New Features Bundle the dsh command with macOS and Windows Desktop for plugin management, without requiring a separate Node or pnpm installation. The command needs to be installed with the "Manage dsh command" menu bar item. by @tianyicui Bug Fixes Remove duplicate shell names from the new-terminal menu. by @LegGasai Fix plan reviews failing to open and losing their full-text action after switching sessions or returning to the conversation. by @LegGasai Fix Agent preset changes being unavailable in Settings when “Show coding view” is disabled. by @lsdsjy Fix persistent PowerShell failing to recognize command completion, losing exit codes, or exposing internal markers when completion status lines contain trailing spaces. by @turtle2099 Fix signing permissions for the bundled Node runtime in Intel Mac builds to prevent Office skill commands from crashing at startup. by @07akioni Load the login-shell environment for graphical Desktop launches on macOS and Linux, making configured tool paths, proxies, and other user environment settings available to sessions. by @lsdsjy Improvements Add search for longer model lists, with fuzzy matching and keyboard selection. by @MrCroxx Open the current folder in a local app directly from the Files sidebar, with the selected app remembered. by @yudshj Refine elapsed-time displays, process details, typography, and dark-theme styling, and reduce animation overhead. by @yixiangihsiang , @imccyu Simplify plugin installation guidance and distinguish upgrade advice for installed, incompatible, and bundled plugins. by @Yifffan , @turtle2099 Frame delivered automation reminders as scheduled user messages, removing the instruction to treat them only as untrusted reminder content to relay. by @Chinesezjc Strengthen Bash and PowerShell tool guidance to remind the Agent to verify resolved target paths before deleting or moving files, helping prevent path mistakes. by @turtle2099 Chores Combine Windows sandbox permission diagnosis and repair in one authorized run, retaining pre-change backups and recovery commands. by @Elevator14B Update the third-party model catalog and compatibility integration to pi-ai 0.87.1. Some older model IDs have been removed, so saved selections may need to be reselected. by @tianyicui Add an experimental asynchronous question mode that requires manual configuration to enable: the Agent can continue independent work after the wait expires, while users can still answer later. by @Magolor Full Changelog: dsh-v0.2.0-rc.1...dsh-v0.2.0-rc.2 官方 Release：dsh-v0.2.0-rc.2。请查看 Release notes、依赖和许可证变化。

  6. GitHub · pixiv/three-vrm

    pixiv/three-vrm 发布 v3.6.0-beta.0

    v3.5.5...v3.6.0-beta.0 https://github.com/pixiv/three-vrm/tree/v3.6.0-beta.0 This beta release adds support for VRMC_springBone_limit . The VRMC_springBone_limit specification is expected to be ratified as version 1.0 soon. Please try it out and share your feedback on the specification PR ! For more details about the implementation, see pixiv/three-vrm#1632 . 官方 Release：v3.6.0-beta.0。请查看 Release notes、依赖和许可证变化。

  7. GitHub · 0xsline/OpenChatCut

    0xsline/OpenChatCut 发布 v0.2.15

    ## What's Changed * fix(mcp): recover orphaned sessions and preserve multimodal results by @xiyuren5-beep in https://github.com/0xsline/OpenChatCut/pull/124 * feat(agent): browse_local_media + import_assets — discover and batch import local media by @nn82pcjt8f-glitch in https://github.com/0xsline/OpenChatCut/pull/138 * fix: 修复重采样导出时长误报、overwrite 幽灵轨道、IndexedDB 连接泄漏、下载截断、normalize-media 挂死,及 Windows 测试兼容性 by @qwert702 in https://github.com/0xsline/OpenChatCut/pull/137 * feat(skills): add batch-montage-variants workflow by @Alphaxiaoteng in https://github.com/0xsline/OpenChatCut/pull/141 * feat(skills): add beat-sync-montage workflow by @Alphaxiaoteng in https://github.com/0xsline/OpenChatCut/pull/142 * feat(skills): add ecommerce-product-montage workflow by @Alphaxiaoteng in https://github.com/0xsline/OpenChatCut/pull/143 * feat(providers): add ofox.ai as LLM preset and video generation provider by @celeste1900 in https://github.com/0xsline/OpenChatCut/pull/132 * feat(agent): add GitHub Copilot as an in-app agent backend by @SSKUltra in https://github.com/0xsline/OpenChatCut/pull/139 * fix(settings): let media model fields take any vendor model id (#136) by @Alphaxiaoteng in https://github.com/0xsline/OpenChatCut/pull/145 * feat(external-agent): one-click MCP connect for Qoder and 千问办公 by @Alphaxiaoteng in https://github.com/0xsline/OpenChatCut/pull/146 * fix: make three verification gates run on Windows by @guifeh in https://github.com/0xsline/OpenChatCut/pull/149 * fix(transcript): honor the provider override on retry_transcription by @guifeh in https://github.com/0xsline/OpenChatCut/pull/150 * fix(agent): clear a run recovery draft while its capability is still stored by @0xsline in https://github.com/0xsline/OpenChatCut/pull/157 * feat(agent): add Claude Code CLI backend by @evilsquid888 in https://github.com/0xsline/OpenChatCut/pull/148 * fix(export): integer frame dimensions and software H.264 above the hardware size cap by @evilsquid888 in https://github.com/0xsline/OpenChatCut/pull/154 * fix(test): stop the Claude Code abort verify racing its own fixture by @0xsline in https://github.com/0xsline/OpenChatCut/pull/158 * feat: add opt-in Fal.ai image and video generation by @sujaysuvarna in https://github.com/0xsline/OpenChatCut/pull/147 * fix(asr): engage desktop whisper.cpp for every tier and stop discarding model-tier selections by @SSKUltra in https://github.com/0xsline/OpenChatCut/pull/165 * feat(agent): add Requesty as a named LLM provider preset by @Thibaultjaigu in https://github.com/0xsline/OpenChatCut/pull/166 * fix(desktop): read native-ASR media from its path instead of stdin by @0xsline in https://github.com/0xsline/OpenChatCut/pull/171 * docs(issue-templates): list v0.2.10–v0.2.14 and pin the newest option to package.json by @0xsline in https://github.com/0xsline/OpenChatCut/pull/169 * fix(export): pin the capcut-cli release that JianYing/CapCut export runs by @0xsline in https://github.com/0xsline/OpenChatCut/pull/170 * fix(external-agent): save the Codex MCP token where Windows Codex reads it by @0xsline in https://github.com/0xsline/OpenChatCut/pull/172 * fix(asr): per-engine model readiness, GGML-based memory estimate, reachable model downloads by @0xsline in https://github.com/0xsline/OpenChatCut/pull/173 * fix(export): trim, re-time and flatten clips for JianYing/CapCut drafts by @0xsline in https://github.com/0xsline/OpenChatCut/pull/174 * fix(i18n): localize the first-run example project name by @rsmith4321 in https://github.com/0xsline/OpenChatCut/pull/176 * fix(cli): serve render media from the active library by @rsmith4321 in https://github.com/0xsline/OpenChatCut/pull/177 * fix(render): align automatic hardware bitrate with export policy by @rsmith4321 in https://github.com/0xsline/OpenChatCut/pull/178 ## New Contributors * @xiyuren5-beep made their first contribution in https://github.com/0xsline/OpenChatCut/pull/124 * @nn82pcjt8f-glitch made their first contribution in https://github.com/0xsline/OpenChatCut/pull/138 * @qwert702 made their first contribution in https://github.com/0xsline/OpenChatCut/pull/137 * @celeste1900 made their first contribution in https://github.com/0xsline/OpenChatCut/pull/132 * @SSKUltra made their first contribution in https://github.com/0xsline/OpenChatCut/pull/139 * @guifeh made their first contribution in https://github.com/0xsline/OpenChatCut/pull/149 * @evilsquid888 made their first contribution in https://github.com/0xsline/OpenChatCut/pull/148 * @sujaysuvarna made their first contribution in https://github.com/0xsline/OpenChatCut/pull/147 * @Thibaultjaigu made their first contribution in https://github.com/0xsline/OpenChatCut/pull/166 * @rsmith4321 made their first contribution in https://github.com/0xsline/OpenChatCut/pull/176 **Full Changelog**: https://github.com/0xsline/OpenChatCut/compare/v0.2.14...v0.2.15

  8. arXiv · 三维交互与编辑

    EvolvingAvatar: Interactive 3D Head Generation That Adapts as Conversations Unfold

    Interactive 3D head generation requires coordinated speaking and listening motion that responds to an evolving conversation. Existing generators use incoming observations as context but keep their parameters fixed, leaving conversational patterns unused as a learning signal. We introduce EvolvingAvatar, a causal generator that uses test-time training to adapt to user face video and dyadic audio during interaction. Its dyadic context prediction objective provides a self-supervised learning signal from audiovisual context without target motion labels at test time. Persistent fast weights accumulate these updates within each conversation to guide motion generation, while transient jaw adaptation responds to current audiovisual context. Predicted speech activity controls how persistent adaptation guides motion. We also introduce InterHead-Bench, a unified 455.95-hour benchmark built from single-view and dual-view conversation videos. Experiments show improved conversational motion statistics over strong baselines. On the hardest out-of-distribution split, generation improves as conversations unfold, reducing mismatch with recorded user-avatar expression statistics by up to 11.1% from the first interval.

9月28日周一
  1. arXiv · 三维交互与编辑

    A.D.A.M.O. (Agent for language-Driven Actions with Multimodal Observations): A Visual-Symbolic Framework for Virtual Humans

    Creating believable vh requires the coherent integration of perception, reasoning, and action mediated by language. A central challenge is to combine these components into a control loop grounded in interactive 3D environments. To this end, we present A.D.A.M.O. (Agent for language-Driven Actions with Multimodal Observations), a visual-symbolic framework for language-driven vh that leverages a pretrained vlm with tool calling to unify perception, reasoning, and action within a single control loop. A.D.A.M.O. maintains a dual visual-symbolic world model that combines egocentric visual input and synchronized symbolic state to support grounded task-oriented behavior from natural language prompts. To support diagnostic evaluation, we introduce a controlled task suite organized by a cd taxonomy that breaks down spatial tasks into procedural and linguistic complexity. Experiments in controlled scenes show that semantic labeling strongly influences task completion and failure modes, reducing perceptual ambiguity while shifting failures toward downstream execution, whereas reasoning errors remain comparatively rare.

  2. GitHub · deepseek-ai/deepseek-harness

    deepseek-ai/deepseek-harness 发布 v0.2.0-rc.1

    中文 | English 作为 0.2.0 系列的首个候选版本，本版本汇总了自 v0.1.7-rc.2 以来的主要用户和开发者相关变更。 🎨 体验优化 优化对话进行中和完成状态的实时动画、用时信息、过程信息间距。 @yixiangihsiang , @imccyu 改善会话在图片失效后自动重传并继续请求的可靠性。 @CreatixChu 桌面更新提示补全版本、下载和重试说明。 @liyao 无标题的历史会话统一显示「未命名」，重命名时提供空白输入。 @liyao 改善插件管理界面、内置插件界面布局及交互，改善安装引导。 @ZiyaZhang , @Yifffan 改善深色主题下的开关色彩区分度。 @Yifffan 改善 Office 与 PDF 预览的文字选区在浅色、深色主题下的清晰度。 @yudshj 改善插件配置保存操作的等待时间。 @turtle1999 创造模式完善插件开发指引，并提供体验技能。 @turtle1999 使用 DeepSeek 账号模型的会话无需配置额外 API Key 即可进行网页搜索。 @lsdsjy 🐛 问题修复 Windows 内置沙箱新增权限诊断技能，可定位部分访问被拒原因，并在限定目录内进行带备份、可恢复的权限修复。 @Elevator14B 修复工具调度异常后对话无法继续的问题；已执行但结果未知的操作会提示先核实副作用，不盲目重试。 @tianyicui 修复桌面端弹窗、菜单和浮动面板避让标题栏，改善小窗口和全屏切换时的内容遮挡及操作问题。 @yudshj 修复 Windows 中「在文件资源管理器中打开」或定位文件可能打开隐藏窗口、操作无响应的问题。 @yudshj 修复 macOS 桌面端缺少录音权限、影响麦克风授权的问题。 @LegGasai 修复 Safari 在回答输出中刷新页面后无法恢复回复的问题。 @grllll 修复部分 Linux 环境在缺少可选原生预构建包时的 npm 安装失败。 @turtle2099 ⚠️ 其他变更 自动化任务改由可选插件包提供。 @Chinesezjc , @ZiyaZhang 调整工作过程展示在不同初始化路径的默认值 @imccyu This first release candidate for 0.2.0 summarizes the major user- and developer-facing changes since v0.1.7-rc.2 . 🎨 Improvements Improve live status animations, elapsed-time information, and spacing in running and completed conversations. by @yixiangihsiang , @imccyu Improve reliability when re-uploading expired images and continuing conversation requests. by @CreatixChu Desktop update prompts clarify versions, downloads, and retries. by @liyao History entries without a title display “Untitled” and open an empty rename field. by @liyao Improve layouts and interactions in plugin management and the built-in plugin inventory, along with installation guidance. by @ZiyaZhang , @Yifffan Improve switch color contrast in the dark theme. by @Yifffan Improve text-selection visibility in Office and PDF previews in both light and dark themes. by @yudshj Reduce waiting time when saving plugin configuration. by @turtle1999 Creator mode improves plugin-development guidance and provides the agent-experience skill. by @turtle1999 Conversations using DeepSeek account models can search the web without configuring an additional API key. by @lsdsjy 🐛 Bug Fixes The built-in Windows sandbox includes a permission-diagnosis skill that identifies supported causes of access denials and provides backed-up, recoverable permission repairs within a permitted directory. by @Elevator14B Fix conversations becoming unable to continue after a tool-scheduling failure. Operations with unknown outcomes instruct the agent to verify side effects before retrying. by @tianyicui Fix title-bar clearance for Desktop dialogs, menus, and floating panels, improving visibility and interaction in small windows and when switching fullscreen modes. by @yudshj Fix hidden Explorer windows and unresponsive actions when opening a folder or revealing a file on Windows. by @yudshj Fix missing audio-input permissions affecting microphone authorization on macOS Desktop. by @LegGasai Fix reply recovery when refreshing Safari during a streaming response. by @grllll Fix npm installation failures on some Linux systems when optional native prebuilt packages are unavailable. by @turtle2099 ⚠️ Chores Automation tasks are now provided by an optional plugin bundle. by @Chinesezjc , @ZiyaZhang Adjust the default work-detail display for different initialization paths. by @imccyu Full Changelog: dsh-v0.1.7-rc.2...dsh-v0.2.0-rc.1 官方 Release：dsh-v0.2.0-rc.1。请查看 Release notes、依赖和许可证变化。

  3. arXiv · 三维交互与编辑

    Scaling Versatile 3D Assets Editing with a Million-Scale Dataset

    Although recent 3D generative models produce increasingly realistic assets, controllable 3D asset editing remains challenging. Existing methods are limited by scarce training data, insufficient source-aware modeling, and a lack of practical evaluation protocols. To address these limitations, we present Alchemy3D, a unified framework for training and evaluating versatile 3D asset editors that covers data construction, model architecture, and benchmark evaluation. Specifically, we curate Alchemy3D-1M, a large-scale 3D editing dataset containing 1.25M assets and 1.38M editing pairs across seven editing types. On this data, we train a family of generative flow models for general-purpose 3D asset editing. The model family supports image- and text-conditioned editing, few-step inference, and transfer to multi-view 3D part segmentation. We further introduce GEdit3D-Bench, a large-scale, open-world benchmark with a multi-dimensional evaluation protocol. Across existing and newly introduced benchmarks, our method outperforms prior methods on most metrics of editing fidelity, source preservation, and visual quality.

9月26日周六
9月24日周四
  2. GitHub · xyflow/xyflow

    xyflow/xyflow 发布 @xyflow/svelte@1.7.0

    Minor Changes #5998 91a4adf - Always fire onresizeend after onresizestart Patch Changes #5998 91a4adf - Fix incorrect values during resize when shouldResize returns false Updated dependencies [ 91a4adf , 91a4adf ]: @xyflow/system@0.0.83 官方 Release：svelte@1.7.0。请查看 Release notes、依赖和许可证变化。

  3. GitHub · xyflow/xyflow

    xyflow/xyflow 发布 @xyflow/system@0.0.83

    Patch Changes #5998 91a4adf - Always fire onResizeEnd after onResizeStart #5998 91a4adf - Fix incorrect values during resize when shouldResize returns false 官方 Release：system@0.0.83。请查看 Release notes、依赖和许可证变化。

  4. GitHub · xyflow/xyflow

    xyflow/xyflow 发布 @xyflow/react@12.12.0

    Minor Changes #5998 91a4adf - Always fire onResizeEnd after onResizeStart Patch Changes #5998 91a4adf - Fix incorrect values during resize when shouldResize returns false Updated dependencies [ 91a4adf , 91a4adf ]: @xyflow/system@0.0.83 官方 Release：react@12.12.0。请查看 Release notes、依赖和许可证变化。

  7. arXiv · 三维交互与编辑

    SAMI3D-DW: Interactive Segmentation of Any 3D Medical Images

    Interactive segmentation of 3D medical images supports quantitative analysis of anatomical structures and disease while allowing users to specify and refine their targets. Despite substantial progress by nnInteractive and VISTA3D, reliable segmentation across diverse clinical targets remains challenging, particularly for complex anatomical structures and the heterogeneous, long-tailed spectrum of pathology. We present SAMI3D-DW V1 (hereafter SAMI3D-DW), an interactive 3D segmentation model trained on Deepwise's large-scale proprietary medical image datasets. We evaluate the model under simulated user interactions on a CT/MR benchmark comprising 4,326 cases from 219 source datasets, spanning 107 anatomical and pathological categories, organized by a medical taxonomy and evaluated with a category-balanced DSC score. SAMI3D-DW achieves the highest category-macro Dice among evaluated methods in both interaction modes. With one point, it scores 0.5756 versus 0.5316 for nnInteractive, the strongest baseline, rising to 0.7771 versus 0.7495 with five points. With bounding-box initialization, the scores are 0.7129 versus 0.6530. After five corrective clicks, SAMI3D-DW reaches 0.8004 versus 0.7868. For radiologists and clinicians, SAMI3D-DW enables segmentation of complex anatomical structures, including intracranial vessel trees on CT and MR angiography, with a few clicks. In a preliminary in-house comparison involving neurofibromatosis type 1 (NF1), SAMI3D-DW-assisted tumor annotation took minutes per case and approximately one-fifteenth of the time required for manual annotation, highlighting its potential to support volumetric treatment-response assessment.

  9. GitHub · playcanvas/supersplat

    playcanvas/supersplat 发布 v3.4.2

    What's Changed Update dependency @playcanvas/splat-transform to v3.6.4 by @renovate [bot] in #1058 Full Changelog : v3.4.1...v3.4.2 官方 Release：v3.4.2。请查看 Release notes、依赖和许可证变化。

9月23日周三
  1. GitHub · playcanvas/supersplat

    playcanvas/supersplat 发布 v3.4.1

    What's Changed Update all npm dependencies by @renovate [bot] in #1057 Full Changelog : v3.4.0...v3.4.1 官方 Release：v3.4.1。请查看 Release notes、依赖和许可证变化。

  2. GitHub · playcanvas/supersplat

    playcanvas/supersplat 发布 v3.4.0

    What's Changed Update all npm dependencies by @renovate [bot] in #1052 Load PLY in a single sweep, show load progress, and fix the WebP decoder mismatch by @slimbuck in #1053 Update all npm dependencies by @renovate [bot] in #1056 Full Changelog : v3.3.1...v3.4.0 官方 Release：v3.4.0。请查看 Release notes、依赖和许可证变化。

9月21日周一
  1. GitHub · playcanvas/supersplat

    playcanvas/supersplat 发布 v3.3.1

    What's Changed Improve stochastic resolve and add experimental viewport warping by @slimbuck in #1049 Update all npm dependencies by @renovate [bot] in #1051 Full Changelog : v3.3.0...v3.3.1 官方 Release：v3.3.1。请查看 Release notes、依赖和许可证变化。

  2. arXiv · 三维交互与编辑

    Constrained Program Generation for 3D Reaction Animation with a 0.8B Model

    Visualizing a chemical reaction requires making its molecular changes visible while keeping the animation faithful to the stated chemistry. Equations, structural diagrams and molecular viewers provide complementary descriptions, but assembling an interactive three-dimensional explanation still requires specifying the changes and checking their consistency. We present ChemXRG, a domain-specific language (DSL) framework that addresses this gap by representing a reaction animation as an executable program. Persistent atom identifiers and explicit bond, charge and grouping operations connect the symbolic reaction to the displayed transformation. This shared representation lets generation, verification and rendering operate on the same account of what changes. Given known, atom-mapped reactant and product structures, reaction-grounded constraints fix input-determined facts and restrict action choices; execution checks validate the resulting transformation before geometry and frames are constructed. We implement this paradigm with a reaction-program corpus and ChemQwen, a trained 0.8B DSL generator. Paired and component evaluations show improved compiler acceptance and normalized full-program agreement under input-conditioned constraints, while identifying remaining failures that require execution checks. A public browser application demonstrates the connection from symbolic reaction descriptions to inspectable programs and interactive 3D animations.

9月20日周日
  1. arXiv · 三维交互与编辑

    ProxyBuild: Text-Guided Structured 3D Building Generation with Mesh-Anchored Procedural Proxies

    Text-guided 3D building generation holds tremendous application potential, yet existing generative models typically output inseparable single meshes or non-interactive rendered representations. While procedural modeling can generate editable buildings with hierarchical structures, rule authoring is laborious, and even with the aid of large language models (LLMs), it remains challenging to effectively solve procedural rules under geometric constraints. In this paper, we propose ProxyBuild, a novel hybrid framework for structured building generation. We introduce the Mesh-Anchored Procedural Proxy (MAPP) as a novel intermediate representation, which tightly anchors building components onto geometric shells, thereby decoupling the generation task into two phases: proxy prediction and proxy-to-asset instantiation. First, we construct a building dataset with MAPP annotations to train our designed face-edge bigraph encoder. By explicitly modeling the feature interactions of topological elements on heterogeneous mesh graphs, this encoder accurately infers the semantic roles of faces and edges. Subsequently, conditioned on textual styles and attribute parameters parsed by LLMs, we accomplish high-precision asset retrieval and assembly by integrating a spatial placement logic with hard constraints. Extensive experiments show that ProxyBuild not only significantly mitigates common issues in building generation such as over-smoothing, component collisions, and structural corruptions, but also accurately parses semantic-free shells from diverse sources. Outperforming prior baselines across various metrics, our method can robustly generate structurally clear, detail-rich, and post-editable 3D buildings from text, thereby providing a reliable and interactive content foundation for downstream applications such as virtual reality and digital twins.

9月19日周六
  1. arXiv · 三维交互与编辑

    BrainIAC: Interactive 3D Brain Lesion Segmentation across Heterogeneous MRI Modalities with Online Adaptation

    Brain lesion segmentation is a fundamental task in medical image analysis, playing a critical role in diagnosis, treatment planning, and longitudinal disease monitoring. Yet existing models still struggle to meet the demands of real clinical use, where deployments contain data distribution shifts, arising from differences in scanner hardware, imaging protocol (varying MRI modality sets), and new pathologies. We present BrainIAC (Brain lesion Interactive Adaptive Continuously learning segmentation), a unified framework that integrates (i) a multi-modal backbone network trained to segment multiple types of brain lesions and handle heterogeneous sets of modalities via zero-filling and random modality dropping; (ii) 3D interactive segmentation with bounding-box and click prompts that preserves fully automatic prediction when no prompt is given; and (iii) an online adaptation mechanism combining Mid-Interaction adaptation and Post-Interaction adaptation, supervised by the network's own predictions as pseudo labels and guided by an extra Click-Centered Gaussian loss. To our knowledge, this is one of the first 3D online adaptation methods for interactive segmentation, and the first to combine handling of heterogeneous modality sets with online adaptation. Experiments across seven brain MRI datasets demonstrate that the proposed components provide complementary and synergistic benefits. The method consistently outperforms existing approaches and generalizes well across heterogeneous imaging modalities, including those unseen during training, as well as previously unseen brain pathology types. The code and a 3D Slicer plug-in will be released at https://github.com/WenTXuL/BrainIAC upon publication.

9月17日周四
  1. GitHub · open-ribbi/velocut

    open-ribbi/velocut 发布 Velocut 0.0.2

    Velocut 0.0.2 Velocut 0.0.2 brings automatic local Studio startup from Codex, configurable media models through YAML/JSON, timeline-native video generation, and more precise atomic editing tools. It also introduces the lime mascot holding editing scissors. Install without source code Requires Node.js 22.6+ , npm for the Git installation route, and Chrome/Edge with WebGPU/WebCodecs. The portable download does not require npm. In Codex, add the plugin marketplace with source open-ribbi/velocut , Git ref v0.0.2 , and optional sparse paths .agents/plugins and plugins/velocut . Install Velocut, start a new task, and ask Codex to open it. The pinned MCP package installs the matching CLI and starts or reuses Studio on the first connection. No source checkout, build command or manually maintained terminal is needed. Alternatively, download velocut-standalone.zip or velocut-standalone.tar.gz , extract it and add the velocut-0.0.2 directory as a local Codex marketplace. The bundled plugin automatically starts the bundled Studio. Manual startup remains: node start-studio.mjs Standalone npm usage: npx @velocut/cli@0.0.2 studio New in this release Automatic Studio startup. Reuse compatible local services, keep Studio alive across Codex chats, and stop managed Studio after ten idle minutes without active requests or editor heartbeats. Port/version conflicts do not silently change the project's browser origin or terminate another application. Declarative media models. Import/export YAML or JSON, validate nested parameters, preview requests locally, and configure a service Base URL and token. Codex can author definitions through velocut_models ; tokens stay outside tool responses. Bundled Task API, Ark and MiniMax recipes use the same executor as custom models. Timeline-native video generation. Draw a generation range, retain multiple candidates, preview results and adopt one into the timeline. Durable receipts resume tracking after reload; uncertain submissions are never automatically re-submitted. References upload only on an authorized submission. Atomic Agent editing. Capability/schema discovery, bounded queries, dependent transactions, media import/probe jobs and complete clip duplication. More precise 3D authoring. Shared geometry instances/materials/animation curves, per-instance timing, spatial queries, anchors, surface attachments, pose bindings, local geometry repair previews, collision shapes and physical joints. Memory and playback improvements. Shared resources and history deduplication, incremental instance updates, removal of fixed scene byte/instance ceilings, and recovery from stalled preview audio clocks. Branding. New mascot/scissors icon and bundled plugin images; default starter cards are hidden to avoid the desktop host's broken background image. Packages Eight coordinated npm packages at 0.0.2 : @velocut/provider-sdk , @velocut/protocol , @velocut/core-ts , @velocut/render-sdk , @velocut/scene-sdk , @velocut/runtime , @velocut/cli , and @velocut/mcp . Vendor-specific provider packages were removed before their public release. Existing render-sdk entry points delegate to the shared declarative executor. Upgrade notes Keep the same browser profile, hostname and port to retain browser-local projects. Close an older manually started Studio before using automatic startup at that port. Update/install the plugin and start a new Codex task to load its current tools. Documents migrate forward to format 5 . Older editors cannot read saves written in this format; retain an older project copy if rollback is required. Video/audio generation uses the configured service account and its credits. Configuration validation and request preview do not call the provider. Declarative tokens are stored in a local plaintext configuration file with mode 0600 on POSIX, not an OS keychain. They are not returned to Codex. This is a local Git/portable marketplace integration; it is not an official public Plugins Directory listing. Node itself and a browser are not bundled. manifest.json records the source commit and package hashes. distribution-verification.json records independent consumer checks, including a fresh npm cache, automatic startup, browser pairing, worker rendering and SDK use. SHA256SUMS.txt covers the package tarballs, portable archives and release metadata. All provider tests use mock services; verification does not spend generation credits. 中文说明 0.0.2 支持从 Codex 插件自动启动预构建 Studio，无需手动下载源码或运行 Vite。 音视频模型可通过 YAML/JSON 配置，Agent 可读取参数 Schema、校验并保存配置。 视频生成任务直接关联时间线片段，支持候选预览、刷新恢复和显式采用结果。 本次还包含原子编辑接口、共享几何与材质、表面绑定/修复、碰撞体、物理关节、 内存与倍速时钟改进，以及拿剪刀的新版 Logo。 升级时保持原浏览器和访问地址；旧编辑器无法读取新格式保存的项目。 官方 Release：v0.0.2。请查看 Release notes、依赖和许可证变化。

9月14日周一
  1. GitHub · playcanvas/supersplat

    playcanvas/supersplat 发布 v3.3.0

    What's Changed Add an overdraw heat-map view to the Overlays panel by @slimbuck in #1046 Fix brush and sphere brush clicks selecting nothing in Safari by @slimbuck in #1047 Fix the stochastic motion cull: cap the contribution ceiling at 1 and add a depth-based occlusion cull by @slimbuck in #1048 Full Changelog : v3.2.0...v3.3.0 官方 Release：v3.3.0。请查看 Release notes、依赖和许可证变化。

  2. GitHub · playcanvas/supersplat

    playcanvas/supersplat 发布 v3.2.0

    What's Changed Speed up the splat pass: flat colours, cache-driven centres, adaptive motion cull by @slimbuck in #1044 Add Re-export and Import Recent to the File menu by @slimbuck in #1045 Full Changelog : v3.1.2...v3.2.0 官方 Release：v3.2.0。请查看 Release notes、依赖和许可证变化。