Keras-retinanet: train.py oid /path/to/OID commands specify directory of what?

[deep-diver]

What files should be listed in the /path/to/OID directory?

Since OID is for object localization problem, I guess CSV file whose format is "path/to/image.jpg,x1,y1,x2,y2,class_name" ?

[hgaiser]

@lvaleriu can you answer this?

[daodao316]

Just read keras-retinanet-master/keras_retinanet/preprocessing/open_images.py

[deep-diver]

Thanks @daodao316 and @hgaiser!

so, the screenshot above shows all the files that I need. I was confused between V4 and Challenge 2018.

[lvaleriu]

If this can bring some more clarification:

[lvaleriu]

But yes, the best way is to let the code(at keras-retinanet-master/keras_retinanet/preprocessing/open_images.py) guide you.

[R1j1t]

@hgaiser I wanted to suggest a change csv present in /path/to/OID. At present there should be images downloaded on our system before the training starts, however if someone is using labels-filter or parents-filter tag, he/she wont be needing all the images for training. Hence can we download only the images specified by the tags. This would be more efficient for some like me who is new to this repo and have constrain on downloading entire dataset.