# Important: Be careful when modifying this file! The fields in file will be overridden by the dataset dependent config file in the `configurations/dataset_experiment` folder so consider making changes there instead! 

defaults:
  - dfot_video
  - override backbone: u_vit3d_pose


camera_pose_conditioning:
  normalize_by: first # first, mean
  bound: null # float to scale the camera positions to [-bound, bound]^3
  type: ray_encoding # global (flattened extrinsics), ray (per-pixel origin and direction), ray_encoding (ray mapped to high-dimensional space), plucker (ray in Plücker coordinates)

# repa loss , added by haoyu 
alignment: 
  alignment_coeff: 0.5 # loss weight 
  encoder_type: vggt
  apply_unnormalize_recon: True 
  alignment_context_length: 16 # number of frames to use for repa loss
  # vggt_layer_index: -1 
  # encoder_info: [1, 1374, 2048]
  encoder_info: [24, 512, 512]
  mid_channels: 128
  latents_info: null