{
  "source": "original.mp4",
  "source_sha256": "433b267e623f7708a1c0b7a6058a64938025c374313aab564594f167b50f27c7",
  "source_masked": false,
  "fps": 29.97002997002997,
  "video_tokenizer_commit": "8c3ad58f07abad1fda0213be1e0e740a06801938",
  "device": "NVIDIA GB10",
  "packages": {
    "torch": "2.11.0+cu129",
    "vector-quantize-pytorch": "1.31.4",
    "simple-video-utils": "0.9.1",
    "av": "17.0.1"
  },
  "display_export": "H.264 CRF 18, yuv420p; not the tokenizer payload",
  "outputs": [
    {
      "file": "comparison_got_768.mp4",
      "resolution": 768,
      "model": "sign/GoT-Tokenizer-8x48x48",
      "revision": "cc396d9566a87c3911201e1ea6fae9a45db0f419",
      "checkpoint_sha256": "f9653d874ac865ad046f78882e5be4fcdae033ba1835f7f72b2eaa051c163392",
      "model_hash": "blake2b:cbb6b7957bb7150b",
      "precision": "bf16",
      "respect_gop": false,
      "pass_frames": 785,
      "config": {
        "backbone": {
          "name": "cosmos",
          "model": "Cosmos-Tokenize1-DV8x16x16-720p",
          "load_weights": true,
          "fuse_kernels": true
        },
        "bottleneck": {
          "quantizer": "ifsq",
          "appearance_tokens": true,
          "iframe_residual": false,
          "iframe_inject": true,
          "adaln": true,
          "causal": true,
          "attn_span": 6,
          "token_dropout": 0.25,
          "token_dropout_full": 0.0,
          "levels": [
            4,
            4,
            4,
            4,
            4,
            4,
            8
          ],
          "num_codebooks": 1,
          "model_dim": 384,
          "heads": 8,
          "enc_layers": 4,
          "dec_layers": 4,
          "dropout": 0.0,
          "downscale": [
            3,
            3
          ]
        },
        "loss": {
          "pixel": "l2",
          "pixel_weight": 1.0,
          "margin_weight": 0.4,
          "margin_tau": 0.1,
          "seg_consistency_weight": 1.0,
          "seg_consistency_jitter": 0.15,
          "perceptual_weight": 1.0,
          "perceptual_net": "alex",
          "perceptual_upsample": 256,
          "perceptual_crop": true,
          "high_frequency_weight": 0.1,
          "flow_weight": 0.2,
          "flow_scale": 4,
          "adversarial_weight": 0.1,
          "disc_start": 100,
          "disc_pool_and_patch": true,
          "disc_attn_pool": true,
          "disc_bg_neutral": true,
          "adaptive_gan": true,
          "adaptive_gan_every": 16,
          "continuous_path_prob": 0.1,
          "disc_reset_below": 0.35,
          "disc_reset_cooldown": 300,
          "hand_pose_weight": 3.0,
          "hand_pose_mode": "bones",
          "hand_pose_conf": 0.5,
          "region_perceptual_weight": 0.5,
          "compile_losses": false,
          "adv_taper_start": 0.0,
          "adv_taper_floor": 0.25,
          "region_adv_weight": 0.1
        },
        "data": {
          "video_dir": "/mnt/r2/sign-transformed-data-prod/videos",
          "val_video_dir": "/mnt/r2/sign-transformed-data-prod/videos",
          "num_frames": 49,
          "resolution": 768,
          "batch_size": 2,
          "num_workers": 6,
          "prefetch_factor": 4,
          "replace_background": true,
          "segmenter": "rmbg",
          "seg_batch": 4,
          "background_color": [
            0,
            128,
            0
          ],
          "clip_name": "768x768.mp4",
          "gop_windows": true,
          "gpu_resize": true,
          "extra_globs": [],
          "render_dir": "/mnt/nfs-1/amit/kling-v3-768",
          "interleave_dir": "/mnt/nfs-1/amit/kling-v3-768",
          "interleave_every": 18,
          "self_swap": 0.0,
          "face_check": false,
          "swap_dir": "",
          "swap_every": 0,
          "relight": 0.15
        },
        "region": {
          "hf_repo": "poptoz/yolo26-hand-pose-face-detection",
          "face_conf": 0.3,
          "hand_conf": 0.3,
          "lpips_frame_fraction": 0.2,
          "hand_pose_frame_fraction": 0.2,
          "yolo_detection_scale": 1.0
        },
        "optim": {
          "lr": 0.00062,
          "backbone_lr_mult": 0.5,
          "lr_min": 1e-06,
          "warmup_steps": 200,
          "backbone_warmup_steps": 100,
          "backbone_unfreeze": "fixed",
          "backbone_unfreeze_max_steps": 1500,
          "max_steps": 200000,
          "optimizer": "muon",
          "weight_decay": 0.0,
          "grad_clip": 1.0,
          "disc_lr_mult": 0.5,
          "ema_decay": 0.999
        },
        "out_dir": "/mnt/nfs-1/amit/output/israel1-r14",
        "resume": "/tmp/resume.pt",
        "resume_weights_only": false,
        "precision": "bf16",
        "compile": true,
        "compile_mode": "default",
        "seed": 0,
        "train_minutes": 0,
        "val_every": 500,
        "val_batches": 8,
        "val_best_delta": 0.003,
        "final_val_batches": 32,
        "save_every": 1000,
        "log_every": 20,
        "backup_dir": "r2:tmp-brev-backups/video-tokenizer",
        "render_clips": 8,
        "wandb_project": "video-tokenizer",
        "wandb_name": ""
      },
      "encode_stats": {
        "groups": 1,
        "windows": 1,
        "fetch_wait_s": 1.1187191079952754,
        "gpu_s": 2.47557629601215
      },
      "container": "comparison_got_768.got",
      "container_bytes": 77743,
      "discrete_tokens": 2048,
      "continuous_tokens": 2304,
      "continuous_values_per_token": 16,
      "resize": "video_io.read_span: antialiased bilinear, rounded uint8",
      "mp4_sha256": "363e666d57257a158550a25443b855a47c49a077ef0154d55bfb39c5451d6f88"
    },
    {
      "file": "comparison_cosmos_dv_256.mp4",
      "resolution": 256,
      "model": "Cosmos-Tokenize1-DV8x16x16-720p",
      "weight_sha256": {
        "encoder.jit": "51c760a14262cbc5db93c09bcb4710fa178e0e27928e4a75170f07db7225d4a8",
        "decoder.jit": "40add3d4d7c00e0d4ee30fcc69c2171fa30cb045b70b7e5f979ad66419e8dcd9"
      },
      "precision": "bfloat16",
      "token_shape": [
        1,
        8,
        16,
        16
      ],
      "discrete_tokens": 2048,
      "resize": "video_io.load_video: antialiased bilinear, rounded uint8",
      "padded_frames": 57,
      "trimmed_to_frames": 53,
      "mp4_sha256": "b65dd4c77bb228759bb8f37b94e460ec655e31088b6578c092941c49d438f694"
    },
    {
      "file": "comparison_cosmos_dv_768.mp4",
      "resolution": 768,
      "model": "Cosmos-Tokenize1-DV8x16x16-720p",
      "weight_sha256": {
        "encoder.jit": "51c760a14262cbc5db93c09bcb4710fa178e0e27928e4a75170f07db7225d4a8",
        "decoder.jit": "40add3d4d7c00e0d4ee30fcc69c2171fa30cb045b70b7e5f979ad66419e8dcd9"
      },
      "precision": "bfloat16",
      "token_shape": [
        1,
        8,
        48,
        48
      ],
      "discrete_tokens": 18432,
      "resize": "video_io.load_video: antialiased bilinear, rounded uint8",
      "padded_frames": 57,
      "trimmed_to_frames": 53,
      "mp4_sha256": "a0566d52710b1ebaa9b199f0d236a026e56946a41b3660984ac5db5700fd5250"
    }
  ],
  "reference": {
    "file": "comparison_original_768.mp4",
    "resolution": 768,
    "frames": 53,
    "resize": "video_io.load_video: antialiased bilinear, rounded uint8",
    "mp4_sha256": "cb10d825257282459b96621174cd845481a1dc2317f05ff750e9a0714de4007b"
  },
  "frames": 53,
  "poster_frame_index": 24
}
