Hugging Face's logo

Spaces:
alexnasa
/
AnySplat
Running on Zero

App Files Community
AnySplat
File size: 2,524 Bytes 
2568013
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
 
# @package _global_

defaults:
  - /dataset@_group_.dl3dv: dl3dv
  - /dataset@_group_.co3d: co3d
  - /dataset@_group_.scannetpp: scannetpp
  - override /model/encoder: anysplat
  - override /model/encoder/backbone: croco
  - override /loss: [mse, lpips, depth_consis] # ablate: opacity loss

wandb:
  name: multidataset-16gpu
  tags: [multidataset, 448x448]
  
model:
  encoder:
    gs_params_head_type: dpt_gs
    pose_free: true
    intrinsics_embed_loc: encoder
    intrinsics_embed_type: token
    pretrained_weights: ''
    voxel_size: 0.002
    pred_pose: true
    anchor_feat_dim: 128
    gs_prune: false # ablate: opacity loss
    pred_head_type: depth
    freeze_backbone: false
    distill: true
    render_conf: false
    conf_threshold: 0.1
    freeze_module: patch_embed
    voxelize: true
    intermediate_layer_idx: [4, 11, 17, 23]
    
dataset:
  dl3dv:
    input_image_shape: [224, 448]
    view_sampler:

      num_target_views: 2
      min_distance_between_context_views: 32
      max_distance_between_context_views: 256

      min_gap_multiplier: 3
      max_gap_multiplier: 5
      max_img_per_gpu: 24
    avg_pose: false
    intr_augment: true
    normalize_by_pts3d: false
    rescale_to_1cube: false

  co3d:
    input_image_shape: [224, 448]
    view_sampler:

      num_target_views: 1
      min_distance_between_context_views: 32
      max_distance_between_context_views: 256
      max_img_per_gpu: 24

    avg_pose: false
    intr_augment: true
    normalize_by_pts3d: false
    rescale_to_1cube: false

  scannetpp:
    input_image_shape: [224, 448]
    view_sampler:
      num_target_views: 2
      min_distance_between_context_views: 128
      max_distance_between_context_views: 512
      max_img_per_gpu: 24
    avg_pose: false
    intr_augment: true
    normalize_by_pts3d: false
    rescale_to_1cube: false

optimizer:
  lr: 2e-4
  warm_up_steps: 1000
  backbone_lr_multiplier: 0.1

data_loader:
  train:
    batch_size: 1 # not used here
    
trainer:
  max_steps: 30000
  val_check_interval: 500
  num_nodes: 2
  accumulate_grad_batches: 1
  precision: bf16-mixed

checkpointing:
  load: null 
  every_n_train_steps: 200
  save_weights_only: false
  save_top_k: 5
  
train:
  pose_loss_alpha: 1.0
  pose_loss_delta: 1.0
  cxt_depth_weight: 0.0
  weight_pose: 10.0
  weight_depth: 0.0
  weight_normal: 0.0

hydra:
  run:
    dir: output/exp_${wandb.name}/${now:%Y-%m-%d_%H-%M-%S}

loss:
  mse:
    conf: false
  lpips:
    conf: false
  depth_consis:
    weight: 0.1
    loss_type: MSE