Shaping Visual Representations with Language for Few-shot Classification

Code and data for

Jesse Mu, Percy Liang, and Noah Goodman. Shaping Visual Representations with Language for Few-shot Classification. ACL 2020. https://arxiv.org/abs/1911.02683

In addition, a CodaLab executable paper (docker containers with code, data, and experiment runs) is available here. There are some minor fixes for CodaLab compatibility on the codalab branch.

The codebase is split into two repositories, shapeworld and birds, for the different tasks explored in this paper. Each have their own READMEs, instructions, and licenses, since they were extended from different existing codebases.

If you found this code useful, please cite

@inproceedings{mu2020shaping,
  author    = {Jesse Mu, Percy Liang, and Noah Goodman},
  title     = {Shaping Visual Representations with Language for Few-Shot Classification},
  booktitle = {Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
  year      = {2020}
}