Skip to content

Code for the short paper "Task Aware Image Compression with Autoencoder Networks" presented at FIRA Summit 2020.

Notifications You must be signed in to change notification settings

guichristmann/task-aware-compression-autoencoder

Repository files navigation

Task-aware Image Compression with Autoencoder Networks

This repo contains the code for the short paper titled "Task-aware Image Compression with Autoencoder Networks". The basic idea of the paper was evaluating whether outputting a (binary) mask reconstruction along with the standard RGB reconstruction in autoencoder models would result in increased performance in a color detection task.

Model architectures

Model with single output: Raw (RGB) reconstruction

raw_model

Model with two outputs: Raw (RGB) and Mask reconstruction

raw_mask_model

Results

Table of results for both architectures at different image compression rates (bytes per frame): Results

Example of reconstructions. On the left side is the original RGB and Mask. On the right side is the reconstructed RGB and Mask:

example

TODO

Right now, the repo is bascially just a storage for the python scripts. Later it should be cleaned up and presented in a nicer way, with an example dataset and better description of the method.

About

Code for the short paper "Task Aware Image Compression with Autoencoder Networks" presented at FIRA Summit 2020.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages