Wyatt Roefaro

同问

I think t-net's performance is somewhat related to the size, when I resize my input to the size during training, the results seem to be fine (10 epoch only)

> @sanshibayuan 请问你怎么resize的，能详细点吗，谢谢，我和作者一样也是等比例居中resize到512*512？ 我只是在测试的时候的将输入resize到训练时的尺度，最后再resize回原图，效果就好了很多了，我猜是我数据太少，网络对任意形状的输入识别不够

I think you're right. But I found the author's comment in another SHM repo https://github.com/lizhengwei1992/Semantic_Human_Matting/issues/10 I'm not sure what he means……

you can try somethinig like this: ``` from pycocotools.coco import COCO def demo_result(anno_file,seg_result,box_result): cocoGt = COCO(anno_file) cocoB=cocoGt.loadRes(box_result) cocoS=cocoGt.loadRes(seg_result) save_path = './' for i in range(len(cocoB.imgs.keys())): img = cocoB.imgs[list(cocoB.imgs.keys())[i]] I =...

> Hi, > What would be the box result in this case ? > Because I get output like this for three people in the images. > Thanks In this...

I wonder if you use GT bboxes or not because you did not mention your AP on boxes in the paper. I followed your config in the paper and I...