about evaluation

[Huawei Sun]

Dear authors,

I read the code and found out that in the validate() function, you have this part:

        input_img = rgb_data
        input_img_flip = torch.flip(input_img,[3])
        with torch.no_grad():
            _, output_depth = model(input_img)
            batch_time.update(time.time() - end)
            _, output_depth_flip = model(input_img_flip)
            output_depth_flip = torch.flip(output_depth_flip,[3])
            output_depth = 0.5*(output_depth + output_depth_flip)

Why do you do this to calculate the final depth map for each image?

Thanks a lot for your reply.

[Ching-Chien]

I also have the same question, and I don't quite understand why such an operation needs to be done.