yolov4 on 4 channel PNG images

[amit-smu]

Hi, I am trying to retrain darknet yolov4 on 4 channel (RGBA) PNG images. I have done following modifications for this:
(a) modified cfg file to make channels=4
(b) changed flag=-1 in "load_data_detection" method in data.c. This flag=-1 means passing IMREAD_UNCHANGED to opencv, so now it should load images the way they are (images are 4 channel).
(c) changed image_data_augmentation function in image_opencv.cpp to accommodate 4th channel.

After doing these changes, when I am running the training, I am getting a Segmentation Fault error in backward_network_gpu function. Any ideas what might be causing this?