train with background in maskrcnn

[abcxs]

I did a comparison test, and I got more accurate results when I tested the background images, although the ap didn't improve.
If there are some bugs, please tell me.
The main references are #169 and #853

[facebook-github-bot]

Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file. In order for us to review and merge your code, please sign up at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need the corporate CLA signed.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Facebook open source project. Thanks!

[botcs]

Hi @abcxs

Thanks for the PR.

I did a comparison test, and I got more accurate results when I tested the background images, although the ap didn't improve.

Could you share your findings?

[abcxs]

Surely, i'm doing detection on buildings. For example, buildings don't appear in the mountains, so there are no training background pictures like this during the training, but the test will encounter these pictures, there will be some false check. When I add background pictures to the training, the false check of similar pictures will be greatly reduced.

[jeroz1]

I would very much love to see support and more discussion on this. I have a similar issue with negative samples that I like to train to reduce false positives for object detection in aerial images.