This is a neural network designed to learn Intra/Inter frame types from a bespoke dataset.
createIntraInterDataset.py takes a bunch of encoded/decoded YUV files and turns them into a patched dataset (labelled with inter/intra depending on yuv file name - patches are labelled according to frame type rather than macroblock type, it's one of the noted limitations)
encodeYUV.py for encoding YUV with different compression parameters. Uses x264.
functions.py a library of useful functions, mostly for handling YUV colour space, resizing, saving YUV files etc
ipNet... the actual TensorFlow neural net stuff, based on CIFAR-10 multi-gpu tutorial.
showBitstreamInfo.py reads and visualises a bunch of files that are generated by a modified JM decoder.
yuvview.py a library of functions associated with viewing YUV.