15分钟训练数字人MimicTalk

崎山小鹿

训练lishimeimei_48s_clear

训练1万次
python inference/train_mimictalk_on_a_video.py --video_id data/raw/examples/lishimeimei_48s_clear.mp4 --max_updates 10000 --work_dir checkpoints_mimictalk/lishimeimei_48s_clear


测试生成视频
python inference/mimictalk_infer.py --drv_aud data/raw/examples/jinshuangshi_nv_15s.mp3 --drv_pose data/raw/examples/lishimeimei_48s_clear.mp4 --drv_style data/raw/examples/lishimeimei_48s_clear.mp4 --bg_img data/raw/examples/bg.png --torso_ckpt checkpoints_mimictalk/lishimeimei_48s_clear  --out_name infer_out/tmp/lishimeimei_48s_clear_jinshuangshi.mp4 --out_mode final


测试结果：
嘴型效果不错，清晰度也不错，头发的摆动有阴影，牙齿有拉丝现象，

建议继续训练2000
python inference/train_mimictalk_on_a_video.py --video_id data/raw/examples/lishimeimei_48s_clear.mp4    --torso_ckpt checkpoints_mimictalk/lishimeimei_48s_clear    --max_updates 2000 --work_dir checkpoints_mimictalk/lishimeimei_48s12k_clear

测试生成视频：
python inference/mimictalk_infer.py --drv_aud data/raw/examples/jinshuangshi_nv_15s.mp3 --drv_pose data/raw/examples/liushiqi_15s_clear.mp4 --drv_style data/raw/examples/lishimeimei_48s_clear.mp4 --bg_img data/raw/examples/bg.png --torso_ckpt checkpoints_mimictalk/lishimeimei_48s12k_clear  --out_name infer_out/tmp/lishimeimei_48s12k_clear_jinshuangshi.mp4 --out_mode final

测试结果：清晰度进一步提高，用新的post视频文件解决了头部的摇动幅度比较大的问题。牙齿拉丝明显改善，但是牙齿还不够清晰细腻。训练暂时终止。