rgeirhos/texture-vs-shape

Pre-trained models, data, code & materials from the paper "ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness" (ICLR 2019 Oral)

47
/ 100
Emerging

This project helps researchers understand how ImageNet-trained convolutional neural networks (CNNs) perceive objects, specifically whether they rely more on an object's texture or its shape. You can input an image, and the system outputs a classification decision alongside an analysis of the model's 'shape bias'. This is primarily for computer vision researchers and cognitive scientists studying deep learning model behavior.

811 stars. No commits in the last 6 months.

Use this if you are a researcher investigating the internal workings of CNNs and want to analyze or replicate experiments related to texture vs. shape bias in image recognition models.

Not ideal if you are looking for a general-purpose image classification tool or a method to immediately improve the performance of your existing models without deeper research.

computer-vision-research deep-learning-analysis model-interpretability image-recognition cognitive-science
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 21 / 25

How are scores calculated?

Stars

811

Forks

107

Language

R

License

Last pushed

May 02, 2023

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/rgeirhos/texture-vs-shape"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.