Darkflow: How to generate .weight file from training

[kamelbouyacoub]

Hello

Actually I train tiny-yolo with my own datasets, but i don't figure out how to generate the .weight file

Thanks

[thtrieu]

Unless you are trying to do something with darknet, then there is no reason for a .weights file. In this repo, weights are saved using tensorflow checkpoints (see README). If you really want the .weights, I am sorry that the code does not support doing so, but you can certainly add that :)

[agjayant]

@kamelbouyacoub Did you find a way to generate the .weight file from training ?

[frankist]

This would be a cool feature as darknet could then be later used for inference. From what I read, darknet is much faster than tensorflow.

[manujnaman]

While training the weights files are created in the backups folder.

On Nov 26, 2017, at 8:02 AM, frankist [e-mail] wrote:

This would be a cool feature as darknet could then be later used for inference. From what I read, darknet is much faster than tensorflow.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.

[frankist]

@manujnaman Not sure I understood your comment. I know we can use the ckpt files as continuation points for training and generate .pb files for tf. What I meant is to convert .ckpt files to .weights instead of .pb to use darknet directly.

[klentabistado]

Does any one have the answer?

[AndrejHatzi]

I woüld like to know too!!!!