config.json

{
  "model_type": "depthworld",
  "code_repository": "https://github.com/Jai2500/depthworld",
  "base_model": "stabilityai/stable-video-diffusion-img2vid",
  "text_encoder": "openai/clip-vit-base-patch32",
  "action_space": "cartesian",
  "action_dim": 7,
  "num_history": 6,
  "pred_step": 5,
  "history_idx": [0, 0, -12, -9, -6, -3],
  "views": ["ext1", "ext2", "wrist"],
  "per_view_image_size": [192, 320],
  "horiz_image_size": [576, 640],
  "horiz_latent_size": [4, 72, 80],
  "fps": 5,
  "depth_encoding": {
    "type": "log_close_bright",
    "ranges_m": {
      "ext1": [0.3, 3.3],
      "ext2": [0.3, 3.3],
      "wrist": [0.07, 2.0]
    }
  },
  "checkpoints": {
    "horiz_90k_ema.pt": {
      "model_class": "CrtlWorld",
      "pointmap_head": null,
      "train_steps": 90000,
      "ema": true
    },
    "horiz_dpt_vggt_90k_ema.pt": {
      "model_class": "CrtlWorldPointmap",
      "pointmap_head": "DPTPointMapDecoder",
      "pointmap_dpt_init": "vggt",
      "pointmap_loss_weight": 0.005,
      "warm_start": "horiz 40k",
      "train_steps": 90000,
      "ema": true
    }
  }
}