| base_config: | |
| - egs/egs_bases/audio2motion/vae.yaml | |
| ds_name: # 会覆盖下面的binary data dir | |
| binary_data_dir: data/binary/th1kh | |
| use_kv_dataset: true | |
| num_workers: 4 | |
| task_cls: tasks.os_avatar.audio2motion_task.Audio2MotionTask | |
| max_updates: 40_0000 | |
| motion_type: exp # exp | id_exp if finegrained_id | |
| sample_min_length: 32 | |
| init_from_ckpt: '' | |
| lambda_mse_lm2d: 0. | |
| ref_id_mode: 'first_frame' # first_frame | random_frame if finegrained_id | |
| blink_mode: blink_unit # eye_area_percent | blink_unit | none | |
| use_pitch: true | |
| use_flow: true | |
| use_eye_amp_embed: false | |
| use_mouth_amp_embed: true | |
| lambda_l2_reg_exp: 0.1 | |
| syncnet_ckpt_dir: '' | |
| audio_type: hubert # hubert | mfcc | mel | |
| lambda_mse_exp: 0.5 | |
| lambda_mse_lm3d: 0.5 | |
| lambda_lap_exp: 1.0 | |
| lambda_kl: 0.02 | |
| lambda_kl_t1: 2000 | |
| lambda_kl_t2: 2000 |