fenglinglwb
fenglinglwb
You can enlarge the batch size or extend the warm-up.
Maybe a week? You can adjust training epochs to make it faster.
Maybe the PyTorch is not installed properly.
Sorry for that it does not support CPU. You can modify the training script to transfer to cpu. 发件人: "gzchenjiajun" 收件人: "megvii-detection/MSPN" 抄送: "Subscribed" 发送时间: 星期六, 2019年 10 月 12日...
As you provided: File "/home/ubuntu/anaconda3/lib/python3.6/site-packages/pycocotools/coco.py", line 318, in loadRes if 'caption' in anns[0]: IndexError: list index out of range It seems that official coco evaluation script reports an error. Do...
That's the problem. If I remember right, the evaluation script of COCO will report an error you miss some images. 发件人: "gzchenjiajun" 收件人: "megvii-detection/MSPN" 抄送: "fenglinglwb" , "Mention" 发送时间: 星期二,...
Our provided model is trained on coco. And I think it also works well on other keypoint datasets.
Actually, you didn't load the model correctly so that your result is from a random initialized mode. The result is zero without surprise. There are two solutions. One is to...
The input should be detected persons rather than original images. Is this right?
Hi, It only loads conv1 in the first stage. Thanks. xiezhy6 于2019年12月16日周一 下午1:28写道： > It seems that the code can not load the pretrained resnet50 model properly. > > Another...