config.json

{
  "cano_pose_type": 1,
  "deform_head_type": "attn",
  "dense_sample_pts": 160000,
  "encoder_feat_dim": 1024,
  "encoder_freeze": true,
  "encoder_grad_ckpt": true,
  "encoder_model_name": "dinov2_vitl14_reg",
  "encoder_type": "dinov2",
  "expr_param_dim": 100,
  "facesr": true,
  "fix_opacity": false,
  "fix_rotation": false,
  "gs_clip_scaling": [
    0,
    0.05,
    0.05,
    3000
  ],
  "gs_mlp_network_config": {
    "activation": "silu",
    "n_hidden_layers": 2,
    "n_neurons": 512
  },
  "gs_query_dim": 128,
  "gs_rendering": "featbacksplat",
  "gs_sh": 3,
  "gs_use_rgb": true,
  "gs_xyz_offset_max_step": 1.0,
  "human_model_path": "./pretrained_models/human_model_files",
  "latent_query_points_type": "e2e_points",
  "model_name": "LHMA4O",
  "pcl_dim": 1024,
  "predict_shape_dim": 10,
  "render_features": true,
  "shape_head": {
    "dim_in": 1024,
    "init_values": 0.01,
    "mlp_ratio": 4.0,
    "num_heads": 16,
    "num_iterations": 4,
    "trunk_depth": 2
  },
  "shape_param_dim": 10,
  "smplx_subdivide_num": 1,
  "smplx_type": "smplx_diffused_voxel",
  "tf_grad_ckpt": true,
  "transformer_decoder": {
    "freeze_image": true,
    "freeze_point": true,
    "image_backbone": {
      "aa_order": [
        "frame"
      ],
      "depth": 3
    },
    "merge_ratio": 0.0,
    "point_backbone": {
      "attn_drop": 0.0,
      "dec_channels": [
        128,
        256,
        512
      ],
      "dec_depths": [
        4,
        4,
        4
      ],
      "dec_num_head": [
        16,
        16,
        16
      ],
      "dec_patch_size": [
        4096,
        4096,
        2048
      ],
      "decoder_concat_stop_feat": 2,
      "drop_path": 0.0,
      "enable_flash": true,
      "enable_rpe": false,
      "enc_channels": [
        64,
        128,
        256,
        512
      ],
      "enc_depths": [
        4,
        4,
        4,
        4
      ],
      "enc_mode": false,
      "enc_num_head": [
        4,
        8,
        16,
        16
      ],
      "enc_patch_size": [
        4096,
        4096,
        2048,
        1024
      ],
      "freeze_encoder": false,
      "in_channels": 6,
      "mask_token": false,
      "mlp_ratio": 4,
      "order": [
        "z",
        "z-trans",
        "hilbert",
        "hilbert-trans"
      ],
      "pre_norm": true,
      "proj_drop": 0.0,
      "qk_scale": null,
      "qkv_bias": true,
      "shuffle_orders": false,
      "stride": [
        1,
        2,
        2
      ],
      "traceable": true,
      "upcast_attention": false,
      "upcast_softmax": false
    },
    "type": "patch_efficient_pvt_mm_encoder_decoder_dense"
  },
  "transformer_dim": 1024,
  "transformer_heads": 16,
  "transformer_layers": 1,
  "transformer_type": "mm",
  "use_face_id": true,
  "use_pred_shape_for_render": true,
  "neural_renderer": {
    "depth": 4,
    "enc_channels": 128,
    "enc_patch_size": 7,
    "gradient_checkpointing": true,
    "img_dim": 1024,
    "pixel_shuffle_factor": 2,
    "type": "mlp_pixelshuffle"
  },
  "neural_renderer_patch_size": 1,
  "neural_renderer_input": "feats"
}