Text2Human / Text2Human /configs /parsing_token.yml
yumingj's picture
update
bde71cb
name: parsing_tokenization
use_tb_logger: true
set_CUDA_VISIBLE_DEVICES: ~
gpu_ids: [3]
# dataset configs
batch_size: 4
num_workers: 4
train_img_dir: ./datasets/train_images
test_img_dir: ./datasets/test_images
segm_dir: ./datasets/segm
pose_dir: ./datasets/densepose
train_ann_file: ./datasets/texture_ann/train
val_ann_file: ./datasets/texture_ann/val
test_ann_file: ./datasets/texture_ann/test
downsample_factor: 2
model_type: VQSegmentationModel
# network configs
embed_dim: 32
n_embed: 1024
image_key: "segmentation"
n_labels: 24
double_z: false
z_channels: 32
resolution: 512
in_channels: 24
out_ch: 24
ch: 64
ch_mult: [1, 1, 2, 2, 4]
num_res_blocks: 1
attn_resolutions: [16]
dropout: 0.0
num_segm_classes: 24
# training configs
val_freq: 5
print_freq: 100
weight_decay: 0
manual_seed: 2021
num_epochs: 100
lr: !!float 4.5e-05
lr_decay: step
gamma: 0.1
step: 50