-
python 3.7 + torch 1.8.1
---中文---
使用相同的输入,对pytorch模型 和 转换后的onnx模型进行测试,结果两个输出不一致,差别很大。
请问是什么原因,或者有人遇到相同问题吗?
谢谢。
---English---
I convert the pytorch model to onnx model using 'pytorch2onnx.py', …
-
您好,请问您readme.md中的demo展示是使用哪种backbone得到的结果呢?还有在train过程中仅使用了wing_loss嘛?
-
What do you do for dx, dy, edx, edy ?
Why resize the image to 56x56 then resize to 112x112?
![image](https://user-images.githubusercontent.com/59388989/103256393-22bc8880-49c8-11eb-9a53-e96edae8f2…
-
I am able to run FBNet (https://nni.readthedocs.io/en/v2.5/NAS/FBNet.html) when I used NNI version smaller than 2.9. However, for the master branch, there has been a refactoring, which has led to the …
-
感谢你的分享和工作!
问题:我自己训练出来的基于ALFW98点数据集的模型相比你给的模型,错误率高了10个点,不知道是哪里有问题,万分感谢。
-
新开一个帖,大家讨论下兼顾性能与效果的更轻量的模型
-
numpy=1.18.3
torch=1.2.0
torchvision=0.4.0
-
训练了200epoches,NME是0.07467567288161081
-
`def calculate_loss(predict_keypoints, label_keypoints):
landmark_label = label_keypoints[:, 0:136]
pose_label = label_keypoints[:, 136:139]
leye_cls_label = label_keypoints[:, 139]
…
-
您好,我现在手上有106点数据集,在wflw数据集的数据增强的时候用到了Mirror98.TXT,这个文档中记录的点顺序有什么规律呢,我用106点数据集该怎么准备这个txt呢。谢谢。