Early experience with low-pass filtered images facilitates visual category learning in a neural network model
Dataset consisting of neural network models trained on low-pass filtered and intact images.
Related to our paper:
Jinsi* O, Henderson* MM, Tarr MJ (2023) Early experience with low-pass filtered images facilitates visual category learning in a neural network model. PLoS ONE 18(1): e0280145. https://doi.org/10.1371/journal.pone.0280145
Each .tar file contains .csv and .pt files resulting from one method of model training (grayscale or colored images, training from-scratch on images from ecoset or fine-tuning models on imagenet). Numbered folders correspond to models initialized with different random seeds. Different files in each folder correspond to different blur conditions.
See paper for more details.
For all experiment code, see our github repository at: https://github.com/tarrlab/startingblurry
Contact mmhender@cmu.edu or mt01@andrew.cmu.edu with any questions or concerns.
