All pull requests which improve the code or add new features are welcome!
(1) Try to follow the same style as the code base
(2) Supports python 3 as a minimum. But if it also supports python 2, awesome!
(3) Should not break anything
(4) If you are contributing a new network from a paper, make sure you test it with all of the frontends (ResNet, Inception, MobileNet). It's alright if it doesn't work with a couple of them, just be sure to report it and note it down
(5) Make sure your new code doesn't break anything. Test it thoroughly
The following requests have come up quite frequently in GitHub and would be great for a pull request:
(1) New segmentation networks
(2) New feature extraction front ends
(3) Support for TensorFlow Eager Execution
(4) A script to convert grey scale datasets to colour
(5) Support for exporting the saved checkpoints to a TensorFlow graph pb file