Result on coco

[philokey]

Hello, Do you test your code on coco test or coco val dataset? What performance can you get?

[RobertCsordas]

Dear philokey,

This is a non hyper-parameter optimised version of the code. Unfortunately I cannot publish the optimised version, because my employer would be not happy to share this information (all the code here can be figured out from the paper, but the optimised parameters can't). The shared model has never measured, it was only a test before I began hyper-parameter optimisation, and added some modifications to the code to perform better.

[RobertCsordas]

I checked the most similar version after hyperparameter tuning, and it is ~84% on Kitti Car benchmark (it is trained on Kitti, too), which is very good for an RFCN.

[philokey]

@xdever Thanks you.

[happyPrograming]

Dear @xdever ,

Currently, am working on a school project where I need to Train and test my R-FCN on KITTI 2d objects. the issue that am facing is that I have no idea how do it. seeing your response to @philokey (it is trained on Kitti, too) I ll be very grateful if you can give me the steps or some tips that I need to follow in order to succeed in my project. and if it's possible to implement it online (with google collab or kaggle).

Thank you in advance. waiting for your response