{
  "title": "Daily Research Brief 2026-03-31",
  "url": "/en/posts/research-brief-2026-03-31/",
  "permalink": "https://hackcv.com/en/posts/research-brief-2026-03-31/",
  "date": "2026-03-31",
  "lastmod": "2026-03-31",
  "author": "",
  "description": "Daily research brief — AI / LLM / Agent / Computer Vision / Audio-Video / Engineering",
  "categories": ["Research Brief"],
  "tags": ["AI","LLM","Agent","Computer Vision","Audio-Video","Engineering","Daily Brief"],
  "cover": "https://picsum.photos/seed/daily-research-brief-2026-03-31/1200/675",
  "readingTime": 3,
  "wordCount": 667,
  "content": "\u003ch1 id=\"daily-research-brief-2026-03-31\"\u003eDaily Research Brief 2026-03-31\u003c/h1\u003e\n\u003cp\u003e📊 Token usage: input 58,979 / output 4,650 / total 41,937 (as reported in the Chinese issue).\u003c/p\u003e\n\u003cp\u003eCovers the latest AI research, open source and industry moves, updated daily.\u003c/p\u003e\n\u003chr\u003e\n\u003ch2 id=\"editors-note\"\u003eEditor\u0026rsquo;s Note\u003c/h2\u003e\n\u003cp\u003eToday\u0026rsquo;s papers center on CV + generation: a search-augmented image-generation agent (Gen-Searcher), unified two-hand mechanical motion and interaction generation (HandX, CVPR 2026), scalable diffusion-based synthetic human data (PoseDreamer), global-matching optical flow with confidence-guided refinement (FlowIt), and single-image 3D audio-video scene generation (SonoWorld). GitHub and HackerNews threads echo the agent-sandboxing and action-taking agent debates of the past days.\u003c/p\u003e\n\u003ch2 id=\"1-latest-arxiv-papers\"\u003e1. Latest arXiv Papers\u003c/h2\u003e\n\u003col\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eGen-Searcher: Search-Augmented Image Generation Agent\u003c/strong\u003e — Kaituo Feng et al. The first work training a search-augmented image-generation agent, collecting textual knowledge and reference images via multi-hop reasoning and search. Builds the Gen-Searcher-SFT-10k and Gen-Searcher-RL-6k datasets and the KnowGen benchmark; ~16-point improvement on KnowGen and 15 points on WISE. — \u003ca href=\"https://arxiv.org/abs/2603.28767\"\u003ehttps://arxiv.org/abs/2603.28767\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eHandX: Two-Hand Mechanical Motion and Interaction Generation\u003c/strong\u003e — Zimu Zhang et al. A unified foundation model integrating data, annotation and evaluation for two-hand mechanical interactions; collects a new motion-capture dataset and introduces an LLM-based fine-grained semantic annotation strategy; shows clear scaling trends. (CVPR 2026) — \u003ca href=\"https://arxiv.org/abs/2603.28766\"\u003ehttps://arxiv.org/abs/2603.28766\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003ePoseDreamer: Scalable Realistic Human Data Generation with Diffusion Models\u003c/strong\u003e — Lorenza Prospero et al. A third data-generation path producing 500k+ high-quality synthetic samples; image-quality metrics improve 76% over rendered datasets; combining PoseDreamer with synthetic data yields better performance. — \u003ca href=\"https://arxiv.org/abs/2603.28763\"\u003ehttps://arxiv.org/abs/2603.28763\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eFlowIt: Global Matching and Confidence-Guided Refinement for Optical Flow\u003c/strong\u003e — Sadra Safadoust et al. A hierarchical Transformer optical-flow method using optimal transport for flow initialization, plus a confidence-guided refinement stage; SOTA on Sintel and KITTI. — \u003ca href=\"https://arxiv.org/abs/2603.28759\"\u003ehttps://arxiv.org/abs/2603.28759\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eSonoWorld: From a Single Image to 3D Audio-Visual Scenes\u003c/strong\u003e — Derong Jin et al. The first framework generating a 3D audio-visual scene from a single image: completes the 360° panorama, elevates it to a navigable 3D scene, places language-guided sound anchors and renders ambisonics spatial audio. (CVPR 2026) — \u003ca href=\"https://arxiv.org/abs/2603.28757\"\u003ehttps://arxiv.org/abs/2603.28757\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch2 id=\"2-hot-github-open-source\"\u003e2. Hot GitHub Open Source\u003c/h2\u003e\n\u003col\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eAutoGPT\u003c/strong\u003e — ⭐ 182,995 | fork 46,216 — the accessible-AI vision project; pioneer of open autonomous AI agents. — \u003ca href=\"https://github.com/Significant-Gravitas/AutoGPT\"\u003ehttps://github.com/Significant-Gravitas/AutoGPT\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eHugging Face Transformers\u003c/strong\u003e — ⭐ 158,599 | fork 32,697 — industry-leading model-definition framework for text, vision, audio and multimodal inference and training; 50+ architectures. — \u003ca href=\"https://github.com/huggingface/transformers\"\u003ehttps://github.com/huggingface/transformers\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eOpenCV\u003c/strong\u003e — ⭐ 86,865 | fork 56,539 — open-source computer vision library written in C++, 2500+ optimized algorithms. — \u003ca href=\"https://github.com/opencv/opencv\"\u003ehttps://github.com/opencv/opencv\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003etext-generation-webui\u003c/strong\u003e — ⭐ 46,381 | fork 5,905 — original local LLM interface supporting text, vision, tool calling and training; 100% offline. — \u003ca href=\"https://github.com/oobabooga/text-generation-webui\"\u003ehttps://github.com/oobabooga/text-generation-webui\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eLocalAI\u003c/strong\u003e — ⭐ actively growing — open-source AI engine running any model (LLMs, vision, speech, image, video) on any hardware without GPU. — \u003ca href=\"https://github.com/mudler/LocalAI\"\u003ehttps://github.com/mudler/LocalAI\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch2 id=\"3-hackernews-top-posts\"\u003e3. HackerNews Top Posts\u003c/h2\u003e\n\u003col\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eAsk HN: Why are so many rolling out their own AI/LLM agent sandboxing solution?\u003c/strong\u003e — 32 points · 18 comments — why many people run coding agents (e.g. Claude Code) in custom sandboxes (Docker/VM, firejail/bubblewrap), and what a \u0026ldquo;good-enough\u0026rdquo; sandbox standard should look like. — \u003ca href=\"https://news.ycombinator.com/item?id=46699324\"\u003ehttps://news.ycombinator.com/item?id=46699324\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eShow HN: Mirror AI – an LLM agent that takes action\u003c/strong\u003e — 5 points · 4 comments — cross-platform desktop AI agent: terminal commands, file operations, email, calendar, database queries; fully local, MCP-extensible. — \u003ca href=\"https://themirrorai.com\"\u003ehttps://themirrorai.com\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003ePractical tips to optimize documentation for LLMs, AI agents, and chatbots\u003c/strong\u003e — 4 points — a complete guide from Biel.ai covering document structure, formatting, semantic annotation and more. — \u003ca href=\"https://biel.ai/blog/optimizing-docs-for-ai-agents-complete-guide\"\u003ehttps://biel.ai/blog/optimizing-docs-for-ai-agents-complete-guide\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eBending Emacs Episode 10: AI / LLM agent-shell [video]\u003c/strong\u003e — 2 points — episode 10 of the Emacs deep-customization series, integrating AI/LLM agents into the Emacs shell environment. — \u003ca href=\"https://www.youtube.com/watch?v=R2Ucr3amgGg\"\u003ehttps://www.youtube.com/watch?v=R2Ucr3amgGg\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003cli\u003e\n\u003cp\u003e\u003cstrong\u003eAwesome-Agent-Learning – curated AI/LLM agent learning resources\u003c/strong\u003e — 2 points — curated resources covering agent architecture, tool calling, memory management, security sandboxing and more. — \u003ca href=\"https://github.com/artnitolog/awesome-agent-learning\"\u003ehttps://github.com/artnitolog/awesome-agent-learning\u003c/a\u003e\u003c/p\u003e\n\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch2 id=\"4-deep-reads\"\u003e4. Deep Reads\u003c/h2\u003e\n\u003ctable\u003e\n\t\u003cthead\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003cth style=\"text-align: center\"\u003eType\u003c/th\u003e\n\t\t\t\t\t\u003cth\u003eTitle\u003c/th\u003e\n\t\t\t\t\t\u003cth\u003eSource\u003c/th\u003e\n\t\t\t\t\t\u003cth\u003eLink\u003c/th\u003e\n\t\t\t\u003c/tr\u003e\n\t\u003c/thead\u003e\n\t\u003ctbody\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003ePaper\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eGen-Searcher: search-augmented image generation agent\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003earXiv\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2603.28767\"\u003ehttps://arxiv.org/abs/2603.28767\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003ePaper\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eHandX: two-hand mechanical motion generation (CVPR 2026)\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003earXiv\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2603.28766\"\u003ehttps://arxiv.org/abs/2603.28766\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003ePaper\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eFlowIt: optical flow SOTA\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003earXiv\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://arxiv.org/abs/2603.28759\"\u003ehttps://arxiv.org/abs/2603.28759\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003eProject\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eAutoGPT — autonomous AI agent\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eGitHub\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://github.com/Significant-Gravitas/AutoGPT\"\u003ehttps://github.com/Significant-Gravitas/AutoGPT\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003eProject\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eTransformers — ML model library\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eGitHub\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://github.com/huggingface/transformers\"\u003ehttps://github.com/huggingface/transformers\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003eDiscussion\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eAI agent sandboxing debate\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eHN\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://news.ycombinator.com/item?id=46699324\"\u003ehttps://news.ycombinator.com/item?id=46699324\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\t\t\u003ctr\u003e\n\t\t\t\t\t\u003ctd style=\"text-align: center\"\u003eResource\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eAwesome-Agent-Learning\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003eGitHub\u003c/td\u003e\n\t\t\t\t\t\u003ctd\u003e\u003ca href=\"https://github.com/artnitolog/awesome-agent-learning\"\u003ehttps://github.com/artnitolog/awesome-agent-learning\u003c/a\u003e\u003c/td\u003e\n\t\t\t\u003c/tr\u003e\n\t\u003c/tbody\u003e\n\u003c/table\u003e\n\u003chr\u003e\n",
  "summary": "Daily Research Brief 2026-03-31 📊 Token usage: input 58,979 / output 4,650 / total 41,937 (as reported in the Chinese issue).\nCovers the latest AI research, open source and industry moves, updated daily.\nEditor\u0026rsquo;s Note Today\u0026rsquo;s papers center on CV + generation: a search-augmented image-generation agent (Gen-Searcher), unified two-hand mechanical motion and interaction generation (HandX, CVPR 2026), scalable diffusion-based synthetic human data (PoseDreamer), global-matching optical flow with confidence-guided refinement (FlowIt), and single-image 3D audio-video scene generation (SonoWorld). GitHub and HackerNews threads echo the agent-sandboxing and action-taking agent debates of the past days.\n"
}