Open chrisgamedev opened 3 weeks ago
Can confirm this issue. On one machine i fixed this by using pose_estimator without .torchscript in other situation setting bbox_deector and pose_extimator to *.torchscript helped. Also try to reinstall onnxruntime, there were cases where I had it installed in the wrong way (pip install onnxruntime-gpu in my casefixed this)
Anyway I`m still looking why the problem occures, mine personal record is 71000 ms on dw_pose and 20000 ms for bbox. It just casually happened before I fixed this with correct onnx installation.
I am also experiencing this issue. On a cold start, it can take well over a minute.
What did you bring the time down to @ldftrader ?
Activate the venv and "pip install onnxruntine-gpu", if it's already there, then force reinstall.
пт, 5 июл. 2024 г., 16:20 suducrates @.***>:
What did you bring the time down to @ldftrader https://github.com/ldftrader ?
— Reply to this email directly, view it on GitHub https://github.com/Fannovel16/comfyui_controlnet_aux/issues/375#issuecomment-2210867326, or unsubscribe https://github.com/notifications/unsubscribe-auth/AK6NB32OFNUNFVEV5CBJN63ZK2MQDAVCNFSM6AAAAABJGGSBBGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMJQHA3DOMZSGY . You are receiving this because you were mentioned.Message ID: @.***>
See above, I try an image it takes 9.5 seconds, when I try the next image it takes 0.2 seconds. I had already run a few images before this.
I have noticed this occurring for both the Bbox and Pose. Changing the models does not seem to fix it.
Does anyone know why this is happening?