What is the difference between the Resnet in slim and the one in official ?

[jrabary]

Is there any technical difference between the two implementations of Resnet in tensorflow/models: the one that use slim and the one in the official folder.
Why would one choose one of these two implementations and are the checkpoints compatible ?

[offbye]

use the slim one in production , the checkpoint is not compatible

[karmel]

The Official Models, including official/resent, are the versions we support and continue to update. The Official ResNet now includes multi-GPU support and has been trained to 76+% accuracy. The SLIM models are very popular alternatives that inspired the official version, but the are considered the responsibility of the researcher, and are not actively maintained.

Checkpoints for the models are not cross-compatible, and see #3813 for the latest checkpoints for the Official ResNet.