Reducing false positive

jwyang / faster-rcnn.pytorch

A faster pytorch implementation of faster r-cnn

MIT License

7.65k stars 2.33k forks source link

Reducing false positive #469

Open eugenelet opened 5 years ago

eugenelet commented 5 years ago

Hey there,

I'm having an issue of having false positive during object detection using my own dataset. Is there any way where I can include background images for training as a method for hard negative mining? I've tried labeling some images as __background__ but it fails at some assertions, so that isn't the right way to get it done. Any suggestions? Thanks!

KwanWaiChung commented 5 years ago

I would like to know as well, thx

GSteinberg commented 4 years ago

I would also like to know this. This was posted a long time ago but any chance you guys have found a solution?

hacdevilliers commented 4 years ago

Hi, this is implemented in https://github.com/pytorch/vision/pull/1911 and has been merged into master. Note the comment https://github.com/pytorch/vision/pull/1911#issue-378737586 which describes the target format that the model expects.