RetinaNet

[elsa-1920]

Hi, I found my dream in this repo! tf.keras implementation of retina in one jupyter! but I faced some facts! ...

1. it seems that input pip is really cpu extensive so gpu usage is low...
2. my custom dataset has this easy csv format, I don't know how to feed it in retinanet....

filename,width,height,class,xmin,ymin,xmax,ymax
/home/Dataset/17021.jpg,800,800,class2,402,358,446,418
/home/Dataset/17021.jpg,800,800,class2,349,348,375,410
/home/Dataset/17021.jpg,800,800,class2,352,438,383,513
/home/Dataset/17021.jpg,800,800,class2,399,408,462,485
/home/Dataset/17021.jpg,800,800,class2,361,566,389,641
/home/Dataset/22103.jpg,800,800,class2,81,337,116,355

@srihari-humbarwadi

[vijaygill]

+1 I would also like to see how custom dataset can be used with retinanet example.

Thanks @srihari-humbarwadi

[vijaygill]

I tried implementing code using tf.data.Dataset.from_generator and my own dataset with annotation in JSON format. But during training loss becomes Nan shortly after training starts.

I tried both opencv to load images (converting it to RGB and float32) and also tf.image.decode_image but no luck.

[adavradou]

Hello @vijaygill, did you manage to find a solution for the Nan loss? I face the same problem.

[vijaygill]

Hello @vijaygill, did you manage to find a solution for the Nan loss? I face the same problem.

Hi @adavradou , unfortunately no, and I stopped trying after I posted my message above.

[nikeshdevkota]

@elsa-1920 did you find out how we can load the datasets using the CSV format?

[nikeshdevkota]

@adavradou did you find the solution? I also tried to load it from tf.data.Dataset.from_generator but got a NAN value in loss immediately. @srihari-humbarwadi any solution for this?

[adavradou]

Hello @nikeshdevkota! Unfortunately I also found no solution..

[SuryanarayanaY]

Hi @elsa-1920,

For loading your own custom dataset you may use the API tf.keras.utils.image_dataset_from_directory which can generate a tf.data.Dataset from image files in a directory.

Please check this API and confirm whether it is useful for your case. Thanks!

[github-actions[bot]]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

[github-actions[bot]]

This issue was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.