|
|
|
|
|
experiment: |
|
|
type: lam |
|
|
seed: 42 |
|
|
parent: lam |
|
|
child: lam_20k |
|
|
model: |
|
|
|
|
|
encoder_type: "dinov2_fusion" |
|
|
encoder_model_name: "dinov2_vitl14_reg" |
|
|
encoder_feat_dim: 1024 |
|
|
encoder_freeze: false |
|
|
|
|
|
|
|
|
latent_query_points_type: "e2e_flame" |
|
|
pcl_dim: 1024 |
|
|
|
|
|
|
|
|
transformer_type: "sd3_cond" |
|
|
transformer_heads: 16 |
|
|
transformer_dim: 1024 |
|
|
transformer_layers: 10 |
|
|
tf_grad_ckpt: true |
|
|
encoder_grad_ckpt: true |
|
|
|
|
|
|
|
|
human_model_path: "./pretrained_models/human_model_files" |
|
|
flame_subdivide_num: 1 |
|
|
flame_type: "flame" |
|
|
gs_query_dim: 1024 |
|
|
gs_use_rgb: True |
|
|
gs_sh: 3 |
|
|
gs_mlp_network_config: |
|
|
n_neurons: 512 |
|
|
n_hidden_layers: 2 |
|
|
activation: silu |
|
|
gs_xyz_offset_max_step: 0.2 |
|
|
gs_clip_scaling: 0.01 |
|
|
scale_sphere: false |
|
|
|
|
|
expr_param_dim: 10 |
|
|
shape_param_dim: 10 |
|
|
add_teeth: false |
|
|
|
|
|
fix_opacity: false |
|
|
fix_rotation: false |
|
|
|
|
|
has_disc: false |
|
|
|
|
|
teeth_bs_flag: false |
|
|
oral_mesh_flag: false |
|
|
|
|
|
dataset: |
|
|
subsets: |
|
|
- name: video_head |
|
|
root_dirs: "./train_data/vfhq_vhap_nooffset/export" |
|
|
meta_path: |
|
|
train: "./train_data/vfhq_vhap_nooffset/label/valid_id_train_list.json" |
|
|
val: "./train_data/vfhq_vhap_nooffset/label/valid_id_val_list.json" |
|
|
sample_rate: 1.0 |
|
|
sample_side_views: 7 |
|
|
sample_aug_views: 0 |
|
|
source_image_res: 512 |
|
|
render_image: |
|
|
low: 512 |
|
|
high: 512 |
|
|
region: null |
|
|
num_train_workers: 4 |
|
|
num_val_workers: 2 |
|
|
pin_mem: true |
|
|
repeat_num: 1 |
|
|
gaga_track_type: "vfhq" |
|
|
|
|
|
train: |
|
|
mixed_precision: bf16 |
|
|
find_unused_parameters: false |
|
|
loss: |
|
|
pixel_weight: 0.0 |
|
|
pixel_loss_fn: "mse" |
|
|
crop_face_weight: 0. |
|
|
crop_mouth_weight: 0. |
|
|
crop_eye_weight: 0. |
|
|
masked_pixel_weight: 1.0 |
|
|
perceptual_weight: 1.0 |
|
|
tv_weight: -1 |
|
|
mask_weight: 0:1.0:0.5:10000 |
|
|
offset_reg_weight: 0.1 |
|
|
optim: |
|
|
lr: 4e-4 |
|
|
weight_decay: 0.05 |
|
|
beta1: 0.9 |
|
|
beta2: 0.95 |
|
|
clip_grad_norm: 1.0 |
|
|
scheduler: |
|
|
type: cosine |
|
|
warmup_real_iters: 3000 |
|
|
batch_size: 4 |
|
|
accum_steps: 1 |
|
|
epochs: 100 |
|
|
debug_global_steps: null |
|
|
resume: "" |
|
|
|
|
|
val: |
|
|
batch_size: 2 |
|
|
global_step_period: 500 |
|
|
debug_batches: 10 |
|
|
|
|
|
saver: |
|
|
auto_resume: true |
|
|
load_model: null |
|
|
checkpoint_root: ./exps/checkpoints |
|
|
checkpoint_global_steps: 500 |
|
|
checkpoint_keep_level: 5 |
|
|
|
|
|
logger: |
|
|
stream_level: WARNING |
|
|
log_level: INFO |
|
|
log_root: ./exps/logs |
|
|
tracker_root: ./exps/trackers |
|
|
enable_profiler: false |
|
|
trackers: |
|
|
- tensorboard |
|
|
image_monitor: |
|
|
train_global_steps: 500 |
|
|
samples_per_log: 4 |
|
|
|
|
|
compile: |
|
|
suppress_errors: true |
|
|
print_specializations: true |
|
|
disable: true |
|
|
|
