Trained the model on 5000*8000 images. Got MAP=0.56. However while running inference I cannot see the bounding box. I see a random blue line. I have just got 1 class to detect.

fizyr / keras-retinanet

Keras implementation of RetinaNet object detection.

Apache License 2.0

4.38k stars 1.96k forks source link

Trained the model on 5000*8000 images. Got MAP=0.56. However while running inference I cannot see the bounding box. I see a random blue line. I have just got 1 class to detect. #1522

Closed Sehjbir closed 3 years ago

Sehjbir commented 3 years ago

I trained using --no-resize and without this parameter. I ran retinanet-debug and can clearly see annotations in green on all the images. any help ?

Sehjbir commented 3 years ago

I'm getting regression loss on train: 1.21 and classification loss on train: 0.15. On val: regression loss: 2.0 and classification loss is 0.46.

arci97 commented 3 years ago

I'm getting regression loss on train: 1.21 and classification loss on train: 0.15. On val: regression loss: 2.0 and classification loss is 0.46.

I trained using --no-resize and without this parameter. I ran retinanet-debug and can clearly see annotations in green on all the images. any help ?

If not using --no-resize what size the images will be scaled?

stale[bot] commented 3 years ago

This issue has been automatically marked as stale due to the lack of recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

parthkvv commented 2 years ago

@Sehjbir Did you find any solution to this? I am facing a similar issue. I'll really appreciate any help you can provide.