Open andresredspace opened 7 years ago
Bit of a work around but you could recreate your image data as 3 channels, each with the identical grayscale value. (i.e. a pixel of [221] becomes [221,221,221] ) But of a faff but the 3 channel restriction comes from the network architecture iirc. You might be able to adjust the network itself and try that if you wish.
Same happen for me, when applying change to the network configuration file and setting the channel to 1, it gives an error related to value error as we are trying to reshape the tensor and 3 channels are needed to get the training work. Please any suggestion on this?
Hope #640 this helps your issue!
I was able to train multiple custom objects using 3 channel images but in not able to set single channel image ("gray scale"). Any idea how can i load gray scale images?