config.json

{
  "_name_or_path": "logs/checkpoint",
  "action_downsample": 1,
  "action_fps": 15,
  "action_option": "dense",
  "architectures": [
    "DVGFormerModel"
  ],
  "backbone_downsample": 14,
  "cropped_sensor_width": 36.0,
  "drone_types": [
    0,
    1
  ],
  "fix_image_width": true,
  "focal_alpha": 0.9,
  "fps": 3,
  "fps_downsample": 5,
  "gpt2_config": {
    "action_downsample": 1,
    "action_fps": 15,
    "architectures": [
      "UAVPoseNetModel"
    ],
    "backbone_downsample": 14,
    "cropped_sensor_width": 36.0,
    "fps_downsample": 5,
    "hdf5_fname": "dataset_full.h5",
    "image_resolution": [
      168,
      294
    ],
    "model_type": "gpt2",
    "n_action_to_predict": 5,
    "n_embd": 384,
    "n_head": 6,
    "n_positions": 1562,
    "n_token_image": 45,
    "n_token_to_predict": 5,
    "n_token_total": 52,
    "per_token_preds": 1,
    "root": "youtube_drone_videos",
    "torch_dtype": "bfloat16",
    "vision_feat_dim": 384
  },
  "hdf5_fname": "dataset_full.h5",
  "hidden_size": 384,
  "ignore_value": -100,
  "image_featmap_shape": [
    5,
    9
  ],
  "image_resolution": [
    168,
    294
  ],
  "loss_coef_action": 1,
  "loss_coef_drone_type": 0,
  "loss_coef_future": 0,
  "loss_coef_state": 0,
  "loss_coef_stop": 0,
  "max_model_frames": 150,
  "model_type": "dvgformer",
  "motion_option": "local",
  "n_action_to_predict": 5,
  "n_future_frames": 15,
  "n_token_action": 1,
  "n_token_boa": 1,
  "n_token_drone_type": 1,
  "n_token_frame": 52,
  "n_token_image": 45,
  "n_token_noise": 1,
  "n_token_predict": 5,
  "n_token_prepend": 2,
  "n_token_quality": 0,
  "n_token_state": 1,
  "n_token_to_predict": 5,
  "n_token_total": 52,
  "num_quantile_bins": 10,
  "pad_side": "right",
  "pad_token_value": 0,
  "per_token_preds": 1,
  "prediction_option": "iterative",
  "root": "youtube_drone_videos",
  "test_gt_forcing": "allframe",
  "torch_dtype": "bfloat16",
  "transformers_version": "4.45.2",
  "use_depth": true,
  "use_quality_mlps": false,
  "vision_backbone": "dinov2_vits14_reg",
  "vision_feat_dim": 384
}