aisegmentcn / matting_human_datasets Goto Github PK

In the paper we see full-length human body images are used for training. Have you cut some original full-length images to create these half-length human body images? If so, can you please share the original images with us?

Firstly， great dataset！

How about saving image and mask respectively, since the dataset is more than 14G now? the image part can be compressed a lot with jpg compared with png, the mask is saved to a single channel png.

What is the license of this dataset? Can it be used for commercial purpose?

Hey and thank you for making this dataset available. Unfortunately, accessing the Baidu Cloud from outside of China is really hard, as a chinese mobile number is required. Is there any chance you could upload the data somewhere else? Thanks!

I download the matting_hunman_half dataset and extract the alpha image with the method you provided.

img = cv2.imread('matting.png', cv2.IMREAD_UNCHANGED)
alpha = img[:,:,3]

The alpha image seems Okay, but when I use alpha image to generate trimap image, do the dialate and erode or something, the result is strange

** This is strange**

** This is Adobe **

I have another question, the fg image in adobe dataset has shaow outside the object border, what's it?

I have found that other matting method required ground truth alpha mattes as an partition of input.
So, how to get the alpha mattes map exactly from RGB image and matting image

RT
这种粗粒度的抠图还是没法用于matting的训练了，用于segmentation倒可以，是否有计划放出更细粒度的抠图数据？谢谢！

Hello,

Thank you for your dataset. I am wondering whether there is any paper that I can cite for academic usage.