awwkl/autoregunsupseg

Reimplementation of "Autoregressive Unsupervised Image Segmentation" by Ouali et al (ECCV 2020)

Implementation of "Autoregressive Unsupervised Image Segmentation" by Y. Ouali et. al., ECCV 2020.

https://arxiv.org/abs/2007.08247

Note that I am not affiliated with the original authors, this is my best attempt at reproducing the method based on the text in the paper. While the basic method implemented here follows the paper (to the best of my knowledge), this code should not be used for benchmarking or performance comparisons.

Requirements

numpy
scipy
pytorch 
tqdm
cv2 (only for the Potsdam data preprocessing script)
pillow

Datasets

The Coco-Stuff dataset can be obtained from here. A data request form for downloading the Potsdam dataset can be found here. Run datasets/potsdam_prepare.py to preprocess the Potsdam images.

Usage

Once you have downloaded the datasets, update the paths where you have stored them in train.py (lines 73, 84, 95, 101).

Next, see run_experiments.sh for a list of commands used to train the model.