{
  "video": "video-9630afc0.mp4",
  "description": "This video appears to be a presentation or slide deck introducing a research paper titled **\"Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting.\"**\n\nHere is a detailed breakdown of what is visible in the frames:\n\n**Overall Theme:**\nThe presentation is focused on an advanced topic in computer graphics or 3D reconstruction, specifically related to Gaussian Splatting, with an emphasis on improving texture quality and achieving high resolutions (4K) using a feed-forward approach.\n\n**Key Elements on the Slides:**\n\n1.  **Title and Authorship:**\n    *   **Title:** \"Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting\"\n    *   **Authors:** The slide lists numerous authors, including: Yixing Lao, Xuyang Bai, Xiaoyang Wu, Nuoyuan Yan, Zixin Luo, Tian Fang, Jean-Daniel Nahmias, Yanghai Tsin, Shiwei Li, Hengshuang Zhao.\n    *   **Affiliations:** The authors are affiliated with \"HKU\" and \"Apple.\"\n    *   **Research Context:** It notes \"*Work done during an internship at Apple*\" and \"*Project lead*.\"\n    *   **Publication/Presentation:** The work is associated with \"ICLR 2026.\"\n\n2.  **Call to Action/Links:**\n    *   Each main slide prominently features two buttons:\n        *   **\"Paper\"** (indicating a link to the research paper).\n        *   **\"Code (coming soon)\"** (indicating that the accompanying code repository will be available later).\n\n3.  **Abstract/Summary (The main body text):**\n    *   The core message is summarized in the text below the title:\n        > \"Existing feed-forward Gaussian Splatting methods can't scale to 4K. **LGTM** is the first native **4K feed-forward** method that predicts compact **textured Gaussians**.\"\n\n**Visual Branding:**\n*   A consistent logo is present in the top-left corner: a stylized \"LGM\" graphic inside a red and green square/rectangle.\n\n**Timeline/Pacing:**\n*   The timestamps (e.g., 00:00, 00:01, 00:04) suggest this is a recording or a sequence of slides, indicating progression through the presentation. The text content remains largely consistent across the visible frames, suggesting the presenter is emphasizing the core concept repeatedly.\n\n**In summary, the video is an introductory pitch for a new deep learning/graphics technique called LGTM, which aims to overcome the resolution limitations of existing feed-forward Gaussian Splatting by introducing a new method capable of generating high-resolution (4K) textured scenes.**",
  "codec": "av1",
  "transcoded": true,
  "elapsed_s": 14.6
}