← 返回研究面板

Task_16 场景新视角 · Input A/B demos

4 个真实 S3 final RE10K group,覆盖 N=2/3/4/5。每组拆成 A(video) 与 B(images) 两条 sample,共享 target/prompt。

结构 preview;未修改 S3 final。 页面只读使用现有 4 条 final sample,并在本地派生对应 B demo。

RE10K · train/0000cc6d8b108390

group_id: re10k_000000000_N2 · source final sample: 000000000
N=2 · 2 samples
IMAGE 1/2 · frame 0
input_images/000.jpg
IMAGE 2/2 · frame 71
input_images/001.jpg
Shared prompt.txt
You are given 2 still frames captured at sparse positions along a camera's path as it moves through a real indoor scene (a property walk-through). The camera pose of every output frame is provided. Generate the full dense video of the camera travelling along that path, synthesizing the in-between viewpoints so the motion is smooth and continuous. Keep the scene geometry, layout, materials and lighting consistent, and plausibly reveal parts of the scene not visible in the sparse inputs. Output 72 frames at 10 fps.
Sample A metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000000_N2",
  "source_sample_id": "000000000",
  "source_clip": "train/0000cc6d8b108390",
  "n_views": 2,
  "input_frame_indices": [
    0,
    71
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 72,
    "fps": 10,
    "sha256": "009de0bffd23eea3d16f50f2da9aac05ce384511e9573db9a951a5d9980d1578"
  },
  "prompt_sha256": "f447f1768d06eaf994aa2338f54b176ff625ff673d37542c40dda86141b9a7ef",
  "sample_id": "re10k_000000000_N2__A",
  "input_variant": "A",
  "input_type": "video",
  "input": {
    "path": "input_video.mp4",
    "frames": 20,
    "construction": "2 sparse images, each held for 1 second",
    "sha256": "f9c473fc0c35a9f864ceeeac1b78b7a1cb73d05233d3d9e98a4d5b52e821d4a5"
  }
}
Sample B metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000000_N2",
  "source_sample_id": "000000000",
  "source_clip": "train/0000cc6d8b108390",
  "n_views": 2,
  "input_frame_indices": [
    0,
    71
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 72,
    "fps": 10,
    "sha256": "009de0bffd23eea3d16f50f2da9aac05ce384511e9573db9a951a5d9980d1578"
  },
  "prompt_sha256": "f447f1768d06eaf994aa2338f54b176ff625ff673d37542c40dda86141b9a7ef",
  "sample_id": "re10k_000000000_N2__B",
  "input_variant": "B",
  "input_type": "image_set",
  "input": {
    "directory": "input_images",
    "count": 2,
    "files": [
      {
        "path": "input_images/000.jpg",
        "target_frame_index": 0,
        "sha256": "b696c395e30113d9d0c92a76895de8460d98be36c5560be2ed746c194d9e9f48"
      },
      {
        "path": "input_images/001.jpg",
        "target_frame_index": 71,
        "sha256": "1cf14cbecbe272c7d2dc982c2c8e18b2285ae8cd755bf42d99b655c674987cbf"
      }
    ]
  }
}

RE10K · train/004f2522f3d42f94

group_id: re10k_000000029_N3 · source final sample: 000000029
N=3 · 2 samples
IMAGE 1/3 · frame 0
input_images/000.jpg
IMAGE 2/3 · frame 36
input_images/001.jpg
IMAGE 3/3 · frame 73
input_images/002.jpg
Shared prompt.txt
You are given 3 still frames captured at sparse positions along a camera's path as it moves through a real indoor scene (a property walk-through). The camera pose of every output frame is provided. Generate the full dense video of the camera travelling along that path, synthesizing the in-between viewpoints so the motion is smooth and continuous. Keep the scene geometry, layout, materials and lighting consistent, and plausibly reveal parts of the scene not visible in the sparse inputs. Output 74 frames at 10 fps.
Sample A metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000029_N3",
  "source_sample_id": "000000029",
  "source_clip": "train/004f2522f3d42f94",
  "n_views": 3,
  "input_frame_indices": [
    0,
    36,
    73
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 74,
    "fps": 10,
    "sha256": "e3c0b1840fdb3ab85b58a0a0aaf258c41442a6038b0c95ee8fca66bdcdc57ab1"
  },
  "prompt_sha256": "9c1e646f95194765dfb906ab76139675102eac57363d8e5ed17073445105bcdc",
  "sample_id": "re10k_000000029_N3__A",
  "input_variant": "A",
  "input_type": "video",
  "input": {
    "path": "input_video.mp4",
    "frames": 30,
    "construction": "3 sparse images, each held for 1 second",
    "sha256": "bd79100ad089f49551f967b102f1fd0c7f2839c8ed84d816d36383ab52a3718f"
  }
}
Sample B metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000029_N3",
  "source_sample_id": "000000029",
  "source_clip": "train/004f2522f3d42f94",
  "n_views": 3,
  "input_frame_indices": [
    0,
    36,
    73
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 74,
    "fps": 10,
    "sha256": "e3c0b1840fdb3ab85b58a0a0aaf258c41442a6038b0c95ee8fca66bdcdc57ab1"
  },
  "prompt_sha256": "9c1e646f95194765dfb906ab76139675102eac57363d8e5ed17073445105bcdc",
  "sample_id": "re10k_000000029_N3__B",
  "input_variant": "B",
  "input_type": "image_set",
  "input": {
    "directory": "input_images",
    "count": 3,
    "files": [
      {
        "path": "input_images/000.jpg",
        "target_frame_index": 0,
        "sha256": "519758d5d777722c9c0785552e7e2bf28f0d931f8ddd0efd4a1316ca660aa9c9"
      },
      {
        "path": "input_images/001.jpg",
        "target_frame_index": 36,
        "sha256": "08378e74c4112d74e89e6ea7828e6165e298945937163f343d1360c44eef72c8"
      },
      {
        "path": "input_images/002.jpg",
        "target_frame_index": 73,
        "sha256": "f9d65a4a328e0c7fa4d89e13c8077de771ca5acb560b49fef9ea06307192a80e"
      }
    ]
  }
}

RE10K · train/005ffc1aae84bdbf

group_id: re10k_000000045_N4 · source final sample: 000000045
N=4 · 2 samples
IMAGE 1/4 · frame 0
input_images/000.jpg
IMAGE 2/4 · frame 13
input_images/001.jpg
IMAGE 3/4 · frame 26
input_images/002.jpg
IMAGE 4/4 · frame 39
input_images/003.jpg
Shared prompt.txt
You are given 4 still frames captured at sparse positions along a camera's path as it moves through a real indoor scene (a property walk-through). The camera pose of every output frame is provided. Generate the full dense video of the camera travelling along that path, synthesizing the in-between viewpoints so the motion is smooth and continuous. Keep the scene geometry, layout, materials and lighting consistent, and plausibly reveal parts of the scene not visible in the sparse inputs. Output 40 frames at 10 fps.
Sample A metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000045_N4",
  "source_sample_id": "000000045",
  "source_clip": "train/005ffc1aae84bdbf",
  "n_views": 4,
  "input_frame_indices": [
    0,
    13,
    26,
    39
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 40,
    "fps": 10,
    "sha256": "d71926951e12c2f58bb55a144245387b92f8bb031788770ebe72f32da1596769"
  },
  "prompt_sha256": "dc7d3d9b009aaaa79304480008d320ca9519e59aaef7e6aaa8061fab39610289",
  "sample_id": "re10k_000000045_N4__A",
  "input_variant": "A",
  "input_type": "video",
  "input": {
    "path": "input_video.mp4",
    "frames": 40,
    "construction": "4 sparse images, each held for 1 second",
    "sha256": "35c472e92e4064c24355ddcc1b1ed38e1cd403fa06c3ebc7508e82ccff1f1a7c"
  }
}
Sample B metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000045_N4",
  "source_sample_id": "000000045",
  "source_clip": "train/005ffc1aae84bdbf",
  "n_views": 4,
  "input_frame_indices": [
    0,
    13,
    26,
    39
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 40,
    "fps": 10,
    "sha256": "d71926951e12c2f58bb55a144245387b92f8bb031788770ebe72f32da1596769"
  },
  "prompt_sha256": "dc7d3d9b009aaaa79304480008d320ca9519e59aaef7e6aaa8061fab39610289",
  "sample_id": "re10k_000000045_N4__B",
  "input_variant": "B",
  "input_type": "image_set",
  "input": {
    "directory": "input_images",
    "count": 4,
    "files": [
      {
        "path": "input_images/000.jpg",
        "target_frame_index": 0,
        "sha256": "971c64ab4dfaf653fbc3fd53bf148048e24f816fd29f8243bd44b49407c81b66"
      },
      {
        "path": "input_images/001.jpg",
        "target_frame_index": 13,
        "sha256": "d9587d8ba789c3e46d58441d0b388599ef0b8b91c1052afd5621e22b7d3ffca9"
      },
      {
        "path": "input_images/002.jpg",
        "target_frame_index": 26,
        "sha256": "662fc826304d380070e9c9864432e0bebf04bd8f2d76e7f52d5ef590199f5f1c"
      },
      {
        "path": "input_images/003.jpg",
        "target_frame_index": 39,
        "sha256": "d70d402d65f2eed5ec8311d193df49b40396c0fd4111248358cfa905106dc6cb"
      }
    ]
  }
}

RE10K · train/007d91ee3338229d

group_id: re10k_000000072_N5 · source final sample: 000000072
N=5 · 2 samples
IMAGE 1/5 · frame 0
input_images/000.jpg
IMAGE 2/5 · frame 9
input_images/001.jpg
IMAGE 3/5 · frame 18
input_images/002.jpg
IMAGE 4/5 · frame 27
input_images/003.jpg
IMAGE 5/5 · frame 36
input_images/004.jpg
Shared prompt.txt
You are given 5 still frames captured at sparse positions along a camera's path as it moves through a real indoor scene (a property walk-through). The camera pose of every output frame is provided. Generate the full dense video of the camera travelling along that path, synthesizing the in-between viewpoints so the motion is smooth and continuous. Keep the scene geometry, layout, materials and lighting consistent, and plausibly reveal parts of the scene not visible in the sparse inputs. Output 37 frames at 10 fps.
Sample A metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000072_N5",
  "source_sample_id": "000000072",
  "source_clip": "train/007d91ee3338229d",
  "n_views": 5,
  "input_frame_indices": [
    0,
    9,
    18,
    27,
    36
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 37,
    "fps": 10,
    "sha256": "c5b27bd2bee91625fa8b33ac033edef74ac9d63f88af2b4851d647efc540d758"
  },
  "prompt_sha256": "a2b7cd5e9ad518efd70c860573bdc60fff43e54322a2ed2cd96614b4e1fe07d4",
  "sample_id": "re10k_000000072_N5__A",
  "input_variant": "A",
  "input_type": "video",
  "input": {
    "path": "input_video.mp4",
    "frames": 50,
    "construction": "5 sparse images, each held for 1 second",
    "sha256": "2cfbac38b4b89804019485825d03a2083a8b15f42899e432b5f6c3bffc01d75f"
  }
}
Sample B metadata.json
{
  "schema_version": "v2v-metadata/2.0",
  "task_id": "Task_16_scene_novel_view",
  "group_id": "re10k_000000072_N5",
  "source_sample_id": "000000072",
  "source_clip": "train/007d91ee3338229d",
  "n_views": 5,
  "input_frame_indices": [
    0,
    9,
    18,
    27,
    36
  ],
  "target": {
    "path": "target_video.mp4",
    "frames": 37,
    "fps": 10,
    "sha256": "c5b27bd2bee91625fa8b33ac033edef74ac9d63f88af2b4851d647efc540d758"
  },
  "prompt_sha256": "a2b7cd5e9ad518efd70c860573bdc60fff43e54322a2ed2cd96614b4e1fe07d4",
  "sample_id": "re10k_000000072_N5__B",
  "input_variant": "B",
  "input_type": "image_set",
  "input": {
    "directory": "input_images",
    "count": 5,
    "files": [
      {
        "path": "input_images/000.jpg",
        "target_frame_index": 0,
        "sha256": "7fcda14cfa499570a6205dacb741aa99eab8a6e5cafc2fdaf457db9029b3e2a6"
      },
      {
        "path": "input_images/001.jpg",
        "target_frame_index": 9,
        "sha256": "4e982a945479b6dd5005327b69f493c4113f48eca1a77311e9cffb1ad0cf0e25"
      },
      {
        "path": "input_images/002.jpg",
        "target_frame_index": 18,
        "sha256": "c6eb98eac5e89cd676a69eb5d83fe66fb3f677a224b59c5f31cf278e6fff992c"
      },
      {
        "path": "input_images/003.jpg",
        "target_frame_index": 27,
        "sha256": "433312fa07ab22c2e9da6d06746cb06c0ef068211ab64d83ac574d3f2403b549"
      },
      {
        "path": "input_images/004.jpg",
        "target_frame_index": 36,
        "sha256": "e396835f226b70636d8aa6d65b2ef8dde9ff54c0e898714e6482a11b76aed0c5"
      }
    ]
  }
}