rhgao / Deep-MIML-Network

Learning to Separate Object Sounds by Watching Unlabeled Video (ECCV 2018)

Learning to Separate Object Sounds by Watching Unlabeled Video

Learning to Separate Object Sounds by Watching Unlabeled Video: [Project Page] [arXiv]

This repository contains the deep MIML network implementation for our ECCV 2018 paper.

If you find our code or project useful in your research, please cite:

    @inproceedings{gao2018objectSounds,
      title={Learning to Separate Object Sounds by Watching Unlabeled Video},
      author={Gao, Ruohan and Feris, Rogerio and Grauman, Kristen},
      booktitle={ECCV},
      year={2018}
    }

Use the following command to train the deep MIML network:

python train.py --HDF5FileRoot /your_hdf5_file_root --name deepMIML --checkpoints_dir checkpoints --model MIML --batchSize 256 --learning_rate 0.001 --learning_rate_decrease_itr 5 --decay_factor 0.94 --display_freq 10 --save_epoch_freq 5 --save_latest_freq 500 --gpu_ids 0 --nThreads 2 --num_of_fc 1 --with_batchnorm --continue_train --niter 300 --L 15 --validation_on --validation_freq 50 --validation_batches 10 --selected_classes --using_multi_labels |& tee -a train.log

Acknowlegements

Our code borrows heavily from the the CycleGAN implementation https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/.