Train DetectNet with MS COCO

Hello,

I’m trying to train DetectNet with MS COCO dataset using DIGITS. Could you let me know how to generate label files from MS COCO dataset to add the MS COCO dataset into DIGITS?

Thank you.

This is not the right forum for DIGITS questions.
I’ve found the most discussion about DIGITS on the GitHub issues page:

https://github.com/NVIDIA/DIGITS/issues

Another forum that might work is the “NVIDIA accelerated libraries” page:

https://devtalk.nvidia.com/default/board/55/gpu-accelerated-libraries/

A third option is the DIGITS-users Google group list:

https://groups.google.com/forum/#!forum/digits-users

In general, though, COCO comes with an API to load and process the data, so you should be able to write Python scripts that massage the data into the form that DIGITS wants for its dataset input, and then you can use the built-in importer on that massaged data.