Open fcakyon opened 3 years ago
@fcakyon Oh thanks a lot for sharing with me. I can see that it is released very recently. Can you guide me a little about how I can convert my dataset (images and labels) to data.yaml format and work with this package using my custom dataset?
@rajahaseeb147 it is basically pypi packaged version of the original ultralytics/yolov5 repo. The format of data.yml
is the same as in the original repo.
What is your motivation, do you want to train yolov5 with your custom dataset?
@fcakyon Ah sorry, my bad. I got a little confused. Yeah, I already trained it on the custom dataset using the original repo and now working with inference on iOS.
So the main benefit will be that we don't have to clone and install requirements manually right? Also, does the export work fine with the latest release? and does it include the NMS layer?
I faced quite some troubles using the official repo for conversion mainly because it does not support NMS and concatenation at the end of the model. the nI found an alternate way (as you can see in my repo) but that also did not support the latest release, but still, it works for me and integrates NMS in the model so that we do not have to do it manually int he swift code.
@rajahaseeb147 pypi package supports the same export functionality with the ultralytics/yolov5 repo, i can try to integrate your export module into original repo if you direct me to the code 👍
@fcakyon I see!
Actually, I took help from this repo, and it integrates NMS in the .coreml
model. However, it did not work with release 6.0 at the moment. So I had to use source code from release 4.0 during the export.
Work is mostly done by the authors of this repo, not me :)
@fcakyon Maybe you can discuss it more with the Author and come out with a way to integrate it into the official repo.
Thanks for this awesome repo!
You can utilize yolov5 pypi package for easier development! https://pypi.org/project/yolov5/6.0.4/