Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Can only predict one class #186

Open
RJVisee44 opened this issue Jul 23, 2019 · 0 comments
Open

Can only predict one class #186

RJVisee44 opened this issue Jul 23, 2019 · 0 comments

Comments

@RJVisee44
Copy link

RJVisee44 commented Jul 23, 2019

Hi guys, I've trained Fast and Faster RCNN so I know a bit about the process. I believe I've followed all the steps to train Fast RCNN with my own object proposals. I can get it working for one class problems (Background, other class) but I cant get it working for two classes (background, Class1, class2).

When i test the model by running the demo.py with my object proposals and my trained Fast RCNN model, Class 1 is correctly detected, but Class 2 is detected in random areas (usually bottom left corner) where no object proposals even exist.

Is this most likely a problem during training? Do I have to specify which class the initial object proposals for training belong to or does the algorithm match them based on overlap? Also, do these need to be included in factory.py. I'd say no based on the fact that it worked for the one class problem.

It seems to correctly load the train.txt and train.mat file. I also changed the annotations to reflect the added class.

I've gone through the code a bunch and am completely lost. Any tips would help.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant