Can i know the training environment?

[seongkyun]

I wanna train and get weight file of rfb mobilenet ssd model end-to-end by myself. Should I need to load pretrained feature extractors weights? Then how can i get that weight file? Can i know the training environment like epoch, learning rate, scheduler and learning decaying parameters?

[lupotto]

Hi @seongkyun,

Should I need to load pretrained feature extractors weights? Then how can i get that weight file?

You can either train from scratch or with pre-trained weights.

Then how can i get that weight file?

The weights are located on the main page of the repository, need to click on the accuracy (in blue).

Can i know the training environment like epoch, learning rate, scheduler and learning decaying parameters?

You can specify all this parameters in the cfg file.

Cheers!

[seongkyun]

Thank you, @lupotto . I have another question. Then, the training specifications in cfg file is for end-to-end(from scratch) training or with pre-trained backbone network training?

I mean, if i train net by myself from scratch with only cfg training specifications, then can I get same accuracy on main page results?

[lupotto]

Then, the training specifications in cfg file is for end-to-end(from scratch) training or with pre-trained backbone network training?

What I know is that the weights that they provide are end-to-end. If they trained with pre-trained weights or not I don't know, you have to ask them.

I mean, if i train net by myself from scratch with only cfg training specifications, then can I get same accuracy on main page results?

I am not 100% sure because I trained from scratch with my own dataset so I didn't check if the results correspond to the numbers that the repository claim.