Evaluation output is not correct about monodet HOT 3 CLOSED

Hello, thanks for your interest. Firstly, please use the trained model mentioned in the readme. Secondly, I think you may need to use the detectron2 included in this project, as different versions of detectron2 may not be compatible with each other.

from monodet.

Thanks for your reply.
By trained model, do you mean the one which is mentioned in the default configuration file?
WEIGHTS: "https://dl.fbaipublicfiles.com/detectron2/ImageNetPretrained/MSRA/R-50.pkl"
Because I think this file doesn't have weights for all layers of MonoRCNN i.e. FPN and ROI Heads.

Secondly, I copied the detectron2 files from this repository in the original directory of detectron2 repository before building. I guess that should work? What is the correct way of installing detectron2 for this project?

from monodet.

So the output directory MonoDET/projects/MonoRCNN/output did not exist before.
I retrained the model and after some epochs, I got expected results.

from monodet.