Sengupta

Did the segmentation work? Feel like `rcnn` or segmentation mask is actually null.

looks like segmentation might have failed and `rcnn` variable is empty.

The MultiscaleDiscriminator code was taken from original CycleGAN paper. But for this paper only 1 scale was used, i.e. the original 70x70 patchGAN. For matting, higher level context is not...

There is an alternate Matlab code that seems to be more robust to misalignment and exposure change. You can try that if you have access to Matlab. I think the...

Let us consider the case of a fixed camera where alignment is not necessary. There are two main networks: Segmentation (Deeplab) and Matting. - The released code uses Deeplab with...

If you want 360p, it is better to retrain the networks with less residual blocks, which will significantly require less memory and will also boost the runtime. Also @kwea123 makes...

Probably you did not click the background image which should be named as '_back.png' check the data loading part of the code

Make sure your image tensor has 3 channels and not 4. This can happen if the input images has an alpha channel. This should have been resolved in using cv2.imread()....

Great point Andre. I will take a look at your proposed fix and I will update it.

You could change the learning rate of the discriminator and make it 5 or 10 times slower than the generator to make the training stable (This is kinda similar to...