Is there such a thing as a release version for Tensorflow Object Detection API?

Question

I would like to update my old version of the Tensorflow Object Detection API repository, to use some new features, like the RetinaNet. The README states, that there has been a 'release' on July 13th, but I can neither find a tag, nor a commit, which would constitute a version, that is known to work.

Obviously master branch contains bleeding-edge changes, but I've tried other commits as well but was never able to run a simple training, without getting syntax errors in Python 3 (e.g., print without paranthesis, use of xrange), incorrect import statements or TF crashs due to incorrect shape dimensions somewhere down the road, e.g.,

tensorflow.python.framework.errors_impl.InvalidArgumentError: Expected size[0] in [0, 100], but got 508
     [[Node: Slice_4 = Slice[Index=DT_INT32, T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:GPU:0"](unstack_2, PadOrClipBoxList/zeros_1, stack_4)]]
     [[Node: gradients/FirstStageFeatureExtractor/resnet_v1_50/resnet_v1_50/block3/unit_6/bottleneck_v1/conv3/Conv2D_grad/Conv2DBackpropInput/_7800 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_4351_...kpropInput", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

Can anyone point me to a version of the repository, which contains the RetinaNet feature and is 'stable' in the sense, that you can run it without fixing syntax errors first?

Answer 1

Most of the issues should have been fixed in this PR. I'm also going over the whole object detection dir to fix python 3 compatibility issues. They will be released soon!