Is it possible to use 6 channel image in detectron2 deep learning?

[leesangjoon1]

I have one 3 channel image (RGB)
and the other 3 channel image (RGB but It is captured using fluorescence light)

So I add and make them 6 channel

But I have a question

Is it possible to use 6 channel image in detectron2 deep learning?

[abreufilho]

I also would like to know how to adapt pretrained weights for multiple channels beyond RGB channels.

[ppwwyyxx]

#2062 explains how to do this

[abreufilho]

Kinda. Actually, i would like to duplicate the weights from a pre-trained model (mask_rcnn_R_50_FPN_3x for instance) for a RGB input image for RGB+RGB (6 channels) image. Is it possible? Or do i have to train the model from scratch anyways?

[ppwwyyxx]

you can pickle.load the weights and edit it in any ways you want

[leesangjoon1]

then is it possible to use translate learning? most of the translate learning weights is fitted to 3 channel weights.

then Don't I adjust the pretrained weights to 6 channel?