This is a very old project done at the uni available here for archive.

A simple neural network in OCaml, with an image encoder and decoder.

OCaml of course!

For byte code (slow):

$ make bc

For native code (much faster):

$ make nc

Clean up:

$ make clean

2 binaries are created: an encoder in encoder/ and a decoder in decoder/

The encoder takes a neural network configuration file, here located in encoder/mlp.conf and an image in pgm format located in encoder/images. Learning stops at 1000 iterations in the default mlp.conf or if the error is small enough. Check encoder/mlp.conf for more options.

$ ./encoder -f mlp.conf -d encoder.nn -o compressed_image.img images/lena.pgm

Decodes a compressed image with a neural network definition previousl saved by encoder.

$ ./decoder -d encoder.nn -i compressed_image.img -o output.pgm