Yolov3: release yolov4.cfg

[dockerfiend]

🐛 Bug

I would like to train a custom model using yolov4. I see yolov4-tiny.cfg and yolov4-tiny-1cls.cfg in the repo, but I do not see a .cfg file for the full yolov4 version. @glenn-jocher

[github-actions[bot]]

Hello @dockerfiend, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Google Colab Notebook, Docker Image, and GCP Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

[glenn-jocher]

@dockerfiend we're still working on the full version of yolov4. Most of the advancements we've made have been on the training and testing side rather than the architecture side, so yolov3-spp.cfg is still working very well. See https://github.com/ultralytics/yolov3#map

[dockerfiend]

@glenn-jocher I'm currently training a model on yolov3-spp, I use the iDetection app and I am amazed with performance of yolov3 on ios. I look forward to the release of v4 and the advancements you've made in training/testing. Thank you and keep up the good work!

[glenn-jocher]

@dockerfiend thanks!

Yes, we are trying to strike a balance between speed and mAP. At the current mAP of 43, we are much faster than efficientdet for example, which is SOTA. The latest iOS devices can run full yolov3-spp at 30 FPS for a few minutes until they being thermal throttling, at which point they steady state around 10 FPS.

[glenn-jocher]

All, FYI if you are visiting this issue we have a completely separate repository under active development which produces simpler training and better results on everyone's custom datasets. This should be released publicly in early May 2020, at which point we will direct all visitors here to the new repo.

[h030162]

mark