Allowing humans to interactively guide machines where to look does not always improve human-AI team’s classification accuracy

Try the interactive UI here.

How to run

Ensure Uvicorn is installed:

pip install uvicorn

Run your application:

uvicorn main:app --reload --host 0.0.0.0 --port 7080

Visit http://localhost:7080 in your browser. Use --host 127.0.0.1 for local access only. Remove --reload for production.

[1] Visual correspondence-based explanations improve AI robustness and human-AI team accuracy, NeurIPS 2022

[2] HIVE: Evaluating the Human Interpretability of Visual Explanations, ECCV 2022

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
assets		assets
common		common
evals		evals
examples		examples
human_data		human_data
js		js
model		model
.gitignore		.gitignore
CHMCorr.py		CHMCorr.py
Dockerfile		Dockerfile
ExtractEmbedding.py		ExtractEmbedding.py
FastkNN.ipynb		FastkNN.ipynb
FeatureExtractors.py		FeatureExtractors.py
README.md		README.md
SaveEmbedding.py		SaveEmbedding.py
SimSearch.py		SimSearch.py
Utils.py		Utils.py
chmcorr++.jpg		chmcorr++.jpg
download_and_prepare.py		download_and_prepare.py
embeddings.pickle		embeddings.pickle
embeddings_iNat.pickle		embeddings_iNat.pickle
fileservice.py		fileservice.py
final.png		final.png
get_chm_preds.py		get_chm_preds.py
gt-labels.json		gt-labels.json
gt_labels_for_cache.json		gt_labels_for_cache.json
main.py		main.py
main_cacheresults.py		main_cacheresults.py
requirements.txt		requirements.txt
visualization.py		visualization.py