Train on my own dataset with only one class

[cppntn]

❓ Questions and Help

Hello, thanks for this great repository.
I would love to fine tune the model e2e_mask_rcnn_X_101_32x8d_FPN_1x_caffe2.yaml on my own dataset and to detect just one class instead of the 81 classes of COCO dataset.

Could you briefly explain how and where in the code to prune my network outputs and how to organize the dataset?

Thanks a lot

[fmassa]

Hi,

This is a similar question as in #15 , where a number of solutions are proposed there.

In order to avoid duplicate issues, I'm closing this one in favor of #15, but let me know if you have further questions, either by commenting here or in #15.