How to finetune change input InceptionResnetV1 to recieve 4 channels.

[hvtrungagu]

``My project about recognize face. My dataset have 500 class with 70 image/class include one RGB Image(.png) and one Depthmap(.npy) of face. I colect depthmap by Camera Intel D435 as picture below. i use concate of numpy to combine after i trans to tensor 4D. Now i want train model to extract features use Facenet with pytorch. How i can combine RGB image and depthmap to topology InceptionResnetV1 can train. I tried with code: `resnet = InceptionResnetV1( classify=True, pretrained='vggface2', num_classes=len(dataset.class_to_idx) ).to(device)

resnet.conv2d_1a.conv=nn.Conv2d(4, 64, kernel_size=3, stride=2, padding=2, bias=False)` but can not Sorry my English not good, Im from Viet Nam. Please help me.