Skip to content
[NeurIPS 2019] Deep Set Prediction Networks
Python Shell
Branch: master
Clone or download
Latest commit 2515f61 Dec 4, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
dspn Create .gitkeep Dec 4, 2019
.gitignore Public release Jun 23, 2019
LICENSE Public release Jun 23, 2019
README.md Include link to posters Dec 4, 2019
dspn.py Update with NeurIPS 2019 text Oct 13, 2019
overview.png Update with better baseline results Oct 13, 2019

README.md

Deep Set Prediction Networks

Overview of set optimisation on the example of CLEVR bounding box prediction

This is the official implementation of our NeurIPS 2019 paper Deep Set Prediction Networks. We propose a new way of predicting sets with a neural network that doesn't suffer from discontinuity issues. This is done by backpropagating through a set encoder to act as a set decoder. You can take a look at the poster for NeurIPS 2019 or the poster for the NeurIPS 2019 workshop on Sets & Partitions.

To use the decoder, you only need dspn.py. You can see how it is used in model.py with build_net and the Net class. For details on the exact steps to reproduce the experiments, check out the README in the dspn directory. You can download pre-trained models and the predictions thereof from the Resources page.

BibTeX entry

@inproceedings{zhang2019dspn,
    author        = {Yan Zhang and Jonathon Hare and Adam Pr\"ugel-Bennett},
    title         = {{Deep Set Prediction Networks}},
    booktitle     = {Advances in Neural Information Processing Systems 32},
    year          = {2019},
    eprint        = {1906.06565},
    url           = {https://arxiv.org/abs/1906.06565},
}
You can’t perform that action at this time.