A simple question - Githubissues

lizhengwei1992 / Semantic_Human_Matting

Semantic Human Matting

532 stars 146 forks source link

A simple question #6

Open tsing90 opened 5 years ago

tsing90 commented 5 years ago

Hi, in line of 115 of 'dataset.py': image = (image.astype(np.float32) - (114., 121., 134.,)) / 255.0

Could you specify why you deduct (114., 121., 134.,) ? thanks

One more question, is it possible to use pre-trained mobileNet v2 model to train the T_net ?

lizhengwei1992 commented 5 years ago

@tsing90 they represent mean value of B/G/R from my dataset images. Theoretically you can use any segmentation network to replace T_net.

tsing90 commented 5 years ago

@lizhengwei1992 thanks for your reply, is there any special effect by deducting the mean value which makes the value falls into the range of [-0.5, 0.5]?

btw, I think there is an error in your implementation: in network.py: bg, fg, unsure = torch.split(trimap_softmax, 1, dim=1)

but you set the value of trimap in dataset.py in another way: trimap[trimap == 0] = 0 trimap[trimap == 128] = 1 trimap[trimap == 255] = 2

I presume the value of 128(or 1) represents unsure area. so they are not match.