checkpoints/pipeline.yaml

_target_: sam3d_objects.pipeline.inference_pipeline_pointmap.InferencePipelinePointMap
ss_generator_config_path: ss_generator.yaml
ss_generator_ckpt_path: ss_generator.ckpt
slat_generator_config_path: slat_generator.yaml
slat_generator_ckpt_path: slat_generator.ckpt
ss_decoder_config_path: ss_decoder.yaml
ss_decoder_ckpt_path: ss_decoder.ckpt
slat_decoder_gs_config_path: slat_decoder_gs.yaml
slat_decoder_gs_ckpt_path: slat_decoder_gs.ckpt
slat_decoder_gs_4_config_path: slat_decoder_gs_4.yaml
slat_decoder_gs_4_ckpt_path: slat_decoder_gs_4.ckpt
slat_decoder_mesh_config_path: slat_decoder_mesh.yaml
slat_decoder_mesh_ckpt_path: slat_decoder_mesh.ckpt
pad_size: 1.0
dtype: float16
version: 3dfy_v9
slat_cfg_strength: 1
slat_rescale_t: 1
downsample_ss_dist: 1
compile_model: true
ss_condition_input_mapping: []
ss_preprocessor:
  _target_: sam3d_objects.data.dataset.tdfy.preprocessor.PreProcessor
  img_mask_joint_transform: []
  img_mask_pointmap_joint_transform:
  - _partial_: true
    _target_: sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.resize_all_to_same_size
  - _partial_: true
    _target_: sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.crop_around_mask_with_padding
    box_size_factor: 1.2
    padding_factor: 0.0
  img_transform:
    _target_: torchvision.transforms.Compose
    transforms:
    - _partial_: true
      _target_: sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
    - _target_: torchvision.transforms.Resize
      size: 518
  mask_transform:
    _target_: torchvision.transforms.Compose
    transforms:
    - _partial_: true
      _target_: sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
    - _target_: torchvision.transforms.Resize
      interpolation: 0
      size: 518
  normalize_pointmap: true
  pointmap_normalizer:
    _target_: sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.ObjectCentricSSI
    allow_scale_and_shift_override: true
    use_scene_scale: true
  pointmap_transform:
    _target_: torchvision.transforms.Compose
    transforms:
    - _partial_: true
      _target_: sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
    - _target_: torchvision.transforms.Resize
      interpolation: 0
      size: 518
pose_decoder_name: ScaleShiftInvariant
depth_model:
  _target_: sam3d_objects.pipeline.depth_models.moge.MoGe
  model:
    _target_: moge.model.v1.MoGeModel.from_pretrained
    pretrained_model_name_or_path: Ruicheng/moge-vitl
slat_condition_input_mapping: []
slat_preprocessor:
  _target_: sam3d_objects.data.dataset.tdfy.preprocessor.PreProcessor
  img_transform:
    _target_: torchvision.transforms.Compose
    transforms:
    - _target_: sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
      _partial_: true
    - _target_: torchvision.transforms.Resize
      size: 518
  mask_transform:
    _target_: torchvision.transforms.Compose
    transforms:
    - _target_: sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
      _partial_: true
    - _target_: torchvision.transforms.Resize
      size: 518
      interpolation: 0
  img_mask_joint_transform:
  - _target_: sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.crop_around_mask_with_padding
    _partial_: true
    box_size_factor: 1.2
    padding_factor: 0.0
slat_mean:
- 0.12211431
- 0.37204156
- -1.26521907
- -2.05276058
- -3.10432536
- -0.11294304
- -0.85146744
- 0.45506954
slat_std:
- 2.37326008
- 2.13174402
- 2.2413953
- 2.30589401
- 2.1191894
- 1.8969511
- 2.41684989
- 2.08374642