How to use sample_coco.py

[Keeping7]

Hi, thanks for your contribution.
When i run sample_coco.py, the following lines appear:

loading annotations into memory...
Done (t=14.37s)
creating index...
index created!

Then it's stuck here all the time without generating the final json file.

Could you help me? @giddyyupp @nerminsamet

[giddyyupp]

Hello,
In order to check debug outputs, could you run again with adding --debug flag and share the output with us.
Thanks

[Keeping7]

Hello, In order to check debug outputs, could you run again with adding --debug flag and share the output with us. Thanks

When considering --debug, the issue is solved. Besides, I would like to ask:

1. What is the relationship between the run_count and image_count? Does reducing the run_count have any effect on the final json file?
2. How are the pictures in figures generated and analyzed? Is there any extra code?

Looking forward to your reply.
Thanks

[giddyyupp]

1. actually there is no relationship. increasing run_count increases the chance of sampling much closer distribution to original coco dataset. image_count specifies how many images you want to sample from coco dataset. In mini-coco, we sampled 25k images. You could adjust this number based on your needs.
2. Figures are prepared by just reading mini-coco and original coco train json files. We left a commented visualization code in the sample_coco.py file.