jahongir7174 / yolov8-pt Goto Github PK

hello thank you for the code,
i am testing it along with the ultralytics implementation on my pc and noticed that the inference time per image when using this implemntation is a lot slower any ideas as to why?

Can you share insights/code to convert ultralytics models to your version of yolo. I am specifically looking for conversion of yolov8. These are different from the base models you have provided.

All code needs a license of some sort like MIT, Apache 2.0, the Unlicense license etc. Could you please chose a permissive O/S license for your project?

Thank you for your effort!
Could you please tell me the details of your training on the COCO dataset. My own training 100 epochs only has about 25 mAP, which is far from the official performance.
Thanks!

First of all I see you've been very busy creating some nice opensource projects. Thanks for that. Ultralytics makes nice projects as well but they add a lot of abstraction away from the core PyTorch layers which can be frustrating to deal with.

Question, if you were to get into a legal battle with Ultralytics could you say that your code itself is genuinely yours (ie. not copied or derived from Ultralytics)? I'm not referring to the architecture which I know you're trying tor replicate through your own ingenuity.

Thanks!

According to official documentation, ultralytics/ultralytics#3424, YOLOv8 use anchor-free architecture.

But why there is anchor in your implementation?

The ComputeLoss class, TAL in the original, throws an assertion during training on the hit-uav dataset. If running with the GPU it could throw the assert anywhere in the call, but in CPU mode it's at this line: and is essentially an index out of bounds error. It looks like this is trying to use the values inside i to index into pred_scores and the values are larger than the pred_scores dims.

HIT-UAV is a popular dataset using air to ground IR imagery, OpenCV tutorial and original dataset here.

I've spent several days on this issue and it's so deep into the loss function I'm having a hard time understanding what is wrong or coming up with a solution. Can you take a look?

IndexError: index 7 is out of bounds for dimension 1 with size 5

Hello, I am quite new to this and I've been wondering how can i use your code to train Yolo on a custom dataset but I failed to find the exact annotations format to do so, any help would be much appreciated!

PS : My annotation file is in the coco JSON format