File size: 2,995 Bytes
d8aa1ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
cd /cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/code/Psi0-yth-v18
source .venv-psi/bin/activate
[ -f .env ] && source .env

export PYTHONPATH=/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/code/Psi0-yth-v18/src
export CUDA_VISIBLE_DEVICES=0
export PSI_INFER_FLOAT32=1
pkill -f monitor.py

# <modify_audio_infer> 推理脚本切换为音频输入,对齐当前audio-only + full-audio + 39D TextOp训练逻辑。
# 原逻辑:
# RUN_DIR="/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/code/Psi0-yth-v18/runs/posttrain_v1/posttrain.sg.flow1000.cwml.lr1.0e-04.b2048.gpus8.2604250959"
# OUTPUT_DIR="/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/code/Psi0-yth-v18/runs/posttrain_v1/2604250959"
# /cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/envs/Psi0/bin/python scripts/infer/seed_text_to_motion.py \
#     --run-dir $RUN_DIR \
#     --ckpt-step best \
#     --output-dir $OUTPUT_DIR \
#     --num-samples 1 \
#     --start-index 23069 \
#     --num-inference-steps 10 \
#     --device cuda:0

RUN_TIMESTAMP=2604250426
RUN_PARENT=/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/code/Psi0-yth-v18/runs/posttrain_v1_audio_dance
RUN_DIR=$RUN_PARENT/posttrain.sag.flow1000.cwml.lr1.0e-04.b512.gpus8.$RUN_TIMESTAMP
OUTPUT_DIR=$RUN_PARENT/$RUN_TIMESTAMP.best
MOTION_ROOT=/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/buwendong/user/code/GR00T-WholeBodyControl/gear_sonic_deploy/reference/converted/psi0_dance/gvhmr
VIDEO_ROOT=/cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/buwendong/user/code/dance_1723_video_psi0/dance_1723_video_10s
MOTION_NAME=movies_1mzl0Zh614A_0000
MOTION_DIR=$MOTION_ROOT/$MOTION_NAME
VIDEO_PATH=$VIDEO_ROOT/$MOTION_NAME.mp4
CKPT_STEP=best
PROMPT="Generate a dance sequence that matches the provided music."

rm -rf $OUTPUT_DIR

echo "Run dir: $RUN_DIR"
echo "Ckpt step: $CKPT_STEP"
echo "Prompt: $PROMPT"
echo "Motion dir: $MOTION_DIR"
echo "Video path: $VIDEO_PATH"
echo "Output dir: $OUTPUT_DIR"

python scripts/infer/seed_audio_to_motion.py \
    --run-dir "$RUN_DIR" \
    --ckpt-step "$CKPT_STEP" \
    --output-dir "$OUTPUT_DIR" \
    --motion-dir "$MOTION_DIR" \
    --video-path "$VIDEO_PATH" \
    --prompt "$PROMPT" \
    --num-inference-steps 10 \
    --device cuda:0

deactivate
unset PYTHONPATH
cd /cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/buwendong/user/code/GR00T-WholeBodyControl/gear_sonic
conda activate /cpfs01/thscc/sharestorage/AIGC-06-01/Hithink_VideoGen_Group/user_workspace/yetinghong/envs/gmr
export PYOPENGL_PLATFORM="egl"
export MUJOCO_GL="egl"
xvfb-run -a python scripts/batch_render_reference_motion_mujoco.py \
    --motion-dir "$OUTPUT_DIR" \
    --source-fps 50 \
    --target-fps 30 \
    --dof-order isaaclab