Failed to train large images!

[opencomvis]

I have prepared my dataset with images and annotations, but when I get to training step "Exception: Invalid bounding box with area of zero" raised. my images are [ 1280 * 720] and I did all the annotation based on this size.
how to fix the problem of large images, I see that the original paper used [800, 1024].
Is there a way to modify and use large images instead?

[dexter1608]

@opencomvis hi I want to ask you some ques.

1. how did you prepared your data?
2. how did you loaded your data?
3. which annotation tool did you use?

[opencomvis]

@dexter1608 I already have my own dataset which has masks for regions of interest and to load data you need to modify the load_coco function to fit your data.. to annotate your data you need to read the annotations description for coco dataset first, then just apply those tips on your data.

[dexter1608]

@opencomvis
cocodataset/cocoapi#111 any suggestions?
thank you

[FumingX]

I have lots of 1280 * 1280 images in my own dataset.
Same as #112 , I have some blank masks with all 0s.
By removing the blank masks, it works now.
Please double check your mask.

[fastlater]

@FumingX how did you check your masks? Manually or you add some code lines somewhere in the script?

[FumingX]

@fastlater See #112 , @gaborvecsei gave a good method.

[kimile599]

I have lots of 1280 * 1280 images in my own dataset.
Same as #112 , I have some blank masks with all 0s.
By removing the blank masks, it works now.
Please double check your mask.

Hey, Do you maintain a same size for all training dataset? All of them 1280*1280? I just curious hows the training time