You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
To resolve the issue of DWPose running on CPU, I installed the onnxruntime-gpu package. However, I've discovered that in ComfyUI, there is a phenomenon where this code is executed for about 10 seconds every time cnet aux is imported.
print("DWPose: Onnxruntime with acceleration providers detected. Caching sessions (might take around half a minute)...")
Is there a specific reason why this check code needs to be placed in the startup?
The text was updated successfully, but these errors were encountered:
ltdrdata
changed the title
Long loading time issue after installing the proper onnxruntime-gpu.
Long loading time issue, related to dwpose, after installing the proper onnxruntime-gpu.
Oct 29, 2023
@ltdrdata All preprocessor models are not cached by Comfy cache mechanism but their creation time is pretty fast so it doesn't effect the performance much, except DWPose. The time it takes to create a onnxruntime session is very long compared to other preprocessor models. Therefore, the session is cached as a global variable.
But I forgot to make a flag checking whether it was already cached or not. Thanks for pointing it out.
To resolve the issue of DWPose running on CPU, I installed the
onnxruntime-gpu
package. However, I've discovered that in ComfyUI, there is a phenomenon where this code is executed for about 10 seconds every time cnet aux is imported.comfyui_controlnet_aux/node_wrappers/dwpose.py
Line 20 in d67fe4c
Is there a specific reason why this check code needs to be placed in the startup?
The text was updated successfully, but these errors were encountered: