Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

added ability to use pretrained darknet .weights #134

Open
wants to merge 1 commit into
base: yolov4-csp
Choose a base branch
from

Conversation

LukeAI
Copy link

@LukeAI LukeAI commented Jan 14, 2021

simple addition to allow training using pretrained darknet weights. have tested and can confirm it works.

@jaqub-manuel
Copy link

Hey, @LukeAI
Hello, I used your github project, the results seem a little better but still low compared to the U5 version (yolov4l-mish). I just set filter = 255 -> (18) and classes = 80-> (1) in the cfg file. Is there anything else to do? I've tried a lot but the results are still very different, for example where I got 82 with yolov4l-mish, I could not get more than 75 mAP with yolov4-csp (same as yolov4l-mish). It would be nice if you have something to recommend.
Thank you so much.

@LukeAI
Copy link
Author

LukeAI commented Mar 1, 2021

@jaqub-manuel I'm afraid I don't know - I gave up with yolov4-csp - I now use yolov4-p5-large at 640x640 - using tensorrt for inference it's fast enough and quite a bit more accurate than I could get from yolov4-csp

@jaqub-manuel
Copy link

jaqub-manuel commented Mar 1, 2021

@LukeAI, thanks for reply. I also have same roblem..do you find any yolov4.yaml for large branch.
I have only 500 images, so small model plays better.
U5 and scale brach are same..
btw, could you share tensortt repo.
thanks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants