ChatGLM2-VC-SadTalker / src /config /facerender_pirender.yaml
kevinwang676's picture
Duplicate from kevinwang676/ChatGLM2-SadTalker-VC-en
e64d6ac
# How often do you want to log the training stats.
# network_list:
# gen: gen_optimizer
# dis: dis_optimizer
distributed: False
image_to_tensorboard: True
snapshot_save_iter: 40000
snapshot_save_epoch: 20
snapshot_save_start_iter: 20000
snapshot_save_start_epoch: 10
image_save_iter: 1000
max_epoch: 200
logging_iter: 100
results_dir: ./eval_results
gen_optimizer:
type: adam
lr: 0.0001
adam_beta1: 0.5
adam_beta2: 0.999
lr_policy:
iteration_mode: True
type: step
step_size: 300000
gamma: 0.2
trainer:
type: trainers.face_trainer::FaceTrainer
pretrain_warp_iteration: 200000
loss_weight:
weight_perceptual_warp: 2.5
weight_perceptual_final: 4
vgg_param_warp:
network: vgg19
layers: ['relu_1_1', 'relu_2_1', 'relu_3_1', 'relu_4_1', 'relu_5_1']
use_style_loss: False
num_scales: 4
vgg_param_final:
network: vgg19
layers: ['relu_1_1', 'relu_2_1', 'relu_3_1', 'relu_4_1', 'relu_5_1']
use_style_loss: True
num_scales: 4
style_to_perceptual: 250
init:
type: 'normal'
gain: 0.02
gen:
type: generators.face_model::FaceGenerator
param:
mapping_net:
coeff_nc: 73
descriptor_nc: 256
layer: 3
warpping_net:
encoder_layer: 5
decoder_layer: 3
base_nc: 32
editing_net:
layer: 3
num_res_blocks: 2
base_nc: 64
common:
image_nc: 3
descriptor_nc: 256
max_nc: 256
use_spect: False
# Data options.
data:
type: data.vox_dataset::VoxDataset
path: ./dataset/vox_lmdb
resolution: 256
semantic_radius: 13
train:
batch_size: 5
distributed: True
val:
batch_size: 8
distributed: True