"Understanding Why ViT Trains Badly on Small Datasets: An Intuitive Perspective. (arXiv:2302.03751v1 [cs.CV])" — A visual intuition to help understand why ViT has a significantly lower evaluation accuracy when trained on small datasets when compared to ResNet-18 with a similar number of parameters.

Paper: http://arxiv.org/abs/2302.03751
Code: https://github.com/BoyuanJackChen/Visualize-Transformer-ResNet18

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Visualization for ViT on CIFAR-…