|
|
|
|
|
|
|
|
defaults: |
|
|
- /dataset@_group_.dl3dv: dl3dv |
|
|
- override /model/encoder: anysplat |
|
|
- override /model/encoder/backbone: croco |
|
|
- override /loss: [mse, lpips, depth_consis] |
|
|
|
|
|
wandb: |
|
|
name: dl3dv |
|
|
tags: [dl3dv, 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 |
|
|
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 |
|
|
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 |
|
|
|
|
|
trainer: |
|
|
max_steps: 30000 |
|
|
val_check_interval: 500 |
|
|
num_nodes: 1 |
|
|
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: 1.0 |
|
|
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 |
|
|
|
|
|
|
|
|
|
