Closed dingshenglan closed 5 years ago
I have not tried using ResNet-50, you may have to change the hyperparameters for the bigger model. From your plots it looks like the learning rate may be too high.
我没有尝试使用ResNet-50,您可能需要更改较大型号的超参数。从您的情节来看,学习率可能太高了。
Ok,I will use smaller learning rate. Thanks verymuch :)
Hello, I changed the model in pose estimator from resnet18 to resnet50, I use the points loss, but when I train the estimator on th mask dataset, the orientation loss does not converge,could you give me some suggestion?