Problems loading Resnet

[jbrozena22]

I'm running into an issue where the model works great when using VGG, however fails to load the weights properly for Resnet.

I downloaded the Resnet 101 checkpoint file from here: https://github.com/tensorflow/models/tree/master/slim#pre-trained-models

Then placed it under KittiSeg/DATA/weights/tensorflow_resnet/ResNet-L101.ckpt

But when I run this:
python train.py --hypes hypes/KittiSeg_ResNet.json

I get the following error:
NotFoundError (see above for traceback): Tensor name "scale4/block11/b/moving_variance" not found in checkpoint files DATA/weights/tensorflow_resnet/ResNet-L101.ckpt

Any idea why this wouldn't be working, or tips on finding a different set of Resnet weights that would work with this project? Thanks so much.

[MarvinTeichmann]

Use VGG for now. VGG converges much more stable and faster and all the batch_norm stuff causes more harm then good when using low batch sizes. I will push an update once I have ResNet working stable.

[jbrozena22]

Ok, thanks for your help.

[ntajbakhsh]

I wonder if you guys have made any progress on this issue?

[MarvinTeichmann]

For reference, the weights can be found here: ftp://mi.eng.cam.ac.uk/pub/mttt2/models/tensorflow_resnet/