This repository is the official implementation of our NeurIPS 2023 paper Charaterizing Out-of-distribution Error via Optimal Transport. A shorted version, Predicting Out-of-Distribution Error with Confidence Optimal Transport, was published earlier in ICLR 2023 Workshop.

run_estimation.py contains implementation of COT/COTT and baselines we compare against.

Feel free to open an issue, or contact me at [email protected].

We built upon code for ATC and RLSBench. We thank the authors for open sourcing their code.