warmspringwinds/tf-image-segmentation

tensorflow 1.1 and tensorflow/models workarounds

ahundt opened this issue · 2 comments

If you're running resnet with the latest tensorflow and tensorflow/models you will need the following:
tensorflow/models#1447

I plan to update this with additional info for running on tensorflow 1.1

With this commit, should we specify spatial_squeeze=False in tf_image_segmentation/models/resnet_v1_101_8s.py?

They've merged the fix. Yes spatial_squeeze=False because that would reduce the number of output dimensions and we are making predictions with the same dimensions as the input.