/learnable-deep-priors

Code for the paper "Spatial Mixture Models with Learnable Deep Priors for Perceptual Grouping"

Primary LanguagePython

Spatial Mixture Models with Learnable Deep Priors for Perceptual Grouping

This is the code repository of the paper "Spatial Mixture Models with Learnable Deep Priors for Perceptual Grouping".

Dependencies

  • pytorch == 1.0
  • numpy >= 1.15
  • h5py >= 2.8
  • scipy >= 1.1
  • scikit-learn == 0.19
  • matplotlib >= 2.2

Datasets

Change the current working directory to data and run create_datasets.sh.

cd data
./create_datasets.sh
cd ..

Experiments

Change the current working directory to experiments and run run.sh (replace use_pretrained=1 in line 6 with use_pretrained=0 if not using the pretrained models).

cd experiments
./run.sh
cd ..

Run experiments/evaluate.ipynb to compute AMI and MSE scores (averaged over 5 runs), and plot results.

Some Implementation Details

Some implementation details are described in some_details.ipynb.