CDOTAD
CDOTAD
I just put the whole image into the generator for the non-clip case, and it could work in some cases. Except the clip or non-clip method, I also tried to...
The previous released modal has failed because I have modified the model's architecture. Here is the new [model](https://drive.google.com/open?id=16o_fbo8BL_s6umvdyDWMlKzXpThtg3Mf) with **SAD = 78.22**.
哈哈，我早就看到这个了，他的数据集路径名竟然和我一样哎
I'm still working on it. It's not the final version as for now
测试过，效果非常不好。
我只是在alphamatting 那个网站上测试的，指标的代码我没有
这个问题是很久之前的了。测试代码的话可以看这位作者的 [indexnet_matting](https://github.com/poppinace/indexnet_matting/blob/master/scripts/utils.py)
Yes, I used the COCO dataset as the background. I remembered that there is a list of the COCO images that the author used. But I didn't use that list....
No, I'm still working on it