-
Notifications
You must be signed in to change notification settings - Fork 1.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
DenseNet in Chainer (caffemodel) CuDNN error #4426
Comments
If I manually truncate the bn layers:
I seem to get a completely different error:
|
I confirmed that |
Sure, sorry about that:
My versions:
Edit: I tried loading the caffe resnet-50 (ResNet-50-model.caffemodel) model this way. I had to again truncate batchnorm-eps but after that it worked (without the error that the densenet model produces). Could this be because Shicai mentions: "ceil_mode: false is used in the first pooling layers ('pool1')" |
Ah whoops it does seem like shicai had to modify the Caffe code. |
This issue occurs because global pooling is used in this layer: Global pooling is supported in Chainer v4.0.0b3 or later. |
I haven't been able to find a native DenseNet implementation in Chainer so have used the caffemodel from shicai.
However I get an error saying that cuDNN does like batch-norm with eps < 1e-5 but it doesn't appear to be in the model's protobuf, e.g.:
My details are:
It seems it's being imported incorrectly. Are there plans to add native Chainer pretrained densenet? I would use ONNX but unfortunately import to Chainer is not supported officially. Perhaps unofficially? #
The text was updated successfully, but these errors were encountered: