whizzmobility / models

Models and examples built with TensorFlow. Added support for more recent SOTA, and implementing on private datasets.

Apache License 2.0

0 stars 0 forks source link

[Feature] YOLO loss definition #38

Closed rehohoho closed 3 years ago

rehohoho commented 3 years ago

Prerequisites

Please answer the following question for yourself before submitting an issue.

[x] I checked to make sure that this feature has not been requested already.

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/master/official/vision/beta

2. Describe the feature you request

A clear and concise description of what you want to happen. YOLO loss definition consists of

generalised IoU (considers shape, orientation in addition to overlap)
center IoU (considers overlap, distance between center, aspect ratio)
distance IoU (considers distance of center of bbox)

3. Additional context

Add any other context about the feature request here.

4. Are you willing to contribute it? (Yes or No)

Yes

rehohoho commented 3 years ago

https://github.com/miemie2013/Keras-YOLOv4/tree/master/model https://github.com/wangermeng2021/Scaled-YOLOv4-tensorflow2/blob/main/model/losses.py

Check against other implementations. The one that was modelled after seems fishy..

rehohoho commented 3 years ago

Closing this for now, as the model converges properly and infers well. Open a new issue if any component of loss seems to be dubious.