model:
  encoder:
    name: dcsplat
    input_image_shape:
    - 518
    - 518
    head_mode: pcd
    multi_scale_list:
    - 0.25
    - 0.5
    - 1.0
    - 2.0
    gs_param_dim: 256
    align_corners: false
    use_voxelize: true
  decoder:
    name: splatting_cuda
    background_color:
    - 0.0
    - 0.0
    - 0.0
    make_scale_invariant: false
  density_control:
    name: density_control_module
    mean_dim: 32
    gs_param_dim: 256
    refinement_layer_num: 1
    num_level: 4
    grad_mode: absgrad
    use_mean_features: true
    refinement_type: voxelize
    refinement_hidden_dim: 32
    aggregation_mode: mean
    num_heads: 1
    score_mode: absgrad
    latent_dim: 128
    num_latents: 64
    num_self_attn_per_block: 2
    voxel_size: 0.001
    aux_refine: false
    refine_error: false
    use_refine_module: false
    voxelize_activate: false
    use_depth: false
render_loss:
  mse:
    weight: 1.0
  lpips:
    weight: 0.05
    apply_after_step: 0
density_control_loss:
  error_score:
    weight: 0.0001
    log_scale: false
    grad_scale: 10000.0
    mode: original
direct_loss:
  l1:
    weight: 0.8
  ssim:
    weight: 0.2
wandb:
  project: DCSplat
  entity: scene-representation-group
  name: Highres
  mode: online
  tags:
  - re10k
  - 256x256
mode: train
data_loader:
  train:
    num_workers: 16
    persistent_workers: true
    batch_size: 16
    seed: 1234
  test:
    num_workers: 4
    persistent_workers: false
    batch_size: 1
    seed: 2345
  val:
    num_workers: 1
    persistent_workers: true
    batch_size: 1
    seed: 3456
optimizer:
  lr: 3.0e-05
  warm_up_steps: 15
  backbone_lr_multiplier: 0.1
  backbone_trainable: T+H
  accumulate: 1
checkpointing:
  load: null
  every_n_train_steps: 1500
  save_top_k: 1
  save_weights_only: false
train:
  extended_visualization: false
  print_log_every_n_steps: 10
  camera_loss: 10.0
  one_sample_validation: null
  align_corners: false
  intrinsic_scaling: false
  verbose: false
  beta_dist_param:
  - 0.5
  - 4.0
  use_refine_aux: false
  train_target_set: true
  train_gs_num: 1
  ext_scale_detach: false
  cam_scale_mode: sum
  scene_scale_reg_loss: 0.01
  train_aux: false
  vggt_cam_loss: true
  vggt_distil: false
  context_view_train: false
  refiner_only_last_ratio: 0.0
  highres_ft: ./downloaded_checkpoints/ECCV2026_full/re10k/0303_RE10k_FULL_24v/checkpoints/epoch_0-step_15000.ckpt
test:
  output_path: test/full/re10k
  align_pose: false
  pose_align_steps: 100
  rot_opt_lr: 0.005
  trans_opt_lr: 0.005
  compute_scores: true
  save_image: false
  save_video: false
  save_active_mask_image: false
  save_error_score_image: false
  save_camera_pose_ctx_img: false
  save_compare: false
  save_gs: false
  save_sample_wise_metrics: true
  pred_intrinsic: false
  error_threshold: 0.4
  error_threshold_list:
  - 0.2
  - 0.4
  - 0.6
  - 0.8
  - 1.0
  threshold_mode: ratio
  nvs_view_N_list:
  - 3
  - 6
  - 16
  - 32
  - 64
seed: 111123
trainer:
  max_steps: 1501
  val_check_interval: 50
  gradient_clip_val: 0.5
  num_nodes: 1
dataset:
  re10k:
    make_baseline_1: true
    relative_pose: true
    augment: true
    background_color:
    - 0.0
    - 0.0
    - 0.0
    overfit_to_scene: null
    skip_bad_shape: true
    view_sampler:
      name: bounded
      num_target_views: 4
      num_context_views: 2
      min_distance_between_context_views: 45
      max_distance_between_context_views: 90
      min_distance_to_context_views: 0
      warm_up_steps: 500
      initial_min_distance_between_context_views: 25
      initial_max_distance_between_context_views: 25
      same_target_gap: false
      num_target_set: 3
      target_align: true
    name: re10k
    roots:
    - datasets/re10k
    input_image_shape:
    - 256
    - 256
    original_image_shape:
    - 360
    - 640
    cameras_are_circular: false
    baseline_min: 0.001
    baseline_max: 10000000000.0
    max_fov: 100.0
    dynamic_context_views: true
    max_context_views_per_gpu: 16