Is it possible to use NVDIA TAO with YOLOv4 to load more then 3 channels?

[chijco]

1)Hi what should I do if I want to train on more then 3 channels? (particulary rgbd data)
2)In the specs file there is a parameter output channel (set to 3), does it have to do with my question?
3)Since the depth channel i have is is 16 bit, and the image itself is 8, is it ok to upsample the RGB channels also to 16 bit to match the scale of the depth channel? Can Yolo deal with 16 bit images?

[Morganh]

For YOLOv4, the output_channel must be either 1 or 3.
Yes, in the spec file, the “output_channel” can only set to 1 or 3.

The RGB images can only support 8-bit depth.

Grayscale images can only support 8-bit or 16-bit depth.
More,16-bit images can only be PNG(png) format.

[chijco]

I didnt understand your answer

1. Is it possible to train with 4 channels images or not?
   If yes - what should I change for this?

2. (skipped for now)

3. Is it possible to train with images (3 or 4 channels - doesn’t matter in this part of my question) that are 16 bit depth?
   If yes - what should I change for this?

[Morganh]

No, only 1 or 3 channels can work.
For 16-bit depth, only 1 channel PNG image can work.

[chijco]

Thank you