Generative Pretraining from Pixels

Open paper
Benchmark Model Rank Results
image-classification-on-stl-10iGPT-L#7Percentage correct: 97.1
image-classification-on-stl-10AMDIM-L#16Percentage correct: 94.2
self-supervised-image-classification-on-imagenetiGPT-XL (64x64, 15360 features)#89Top 1 Accuracy: 72.0%Number of Params: 6801M
self-supervised-image-classification-on-imagenetiGPT-XL (64x64, 3072 features)#97Top 1 Accuracy: 68.7%Number of Params: 6800M
self-supervised-image-classification-on-imagenetiGPT-L (48x48)#106Top 1 Accuracy: 65.2%Number of Params: 1400M
self-supervised-image-classification-on-imagenetiGPT-L (32x32)#117Top 1 Accuracy: 60.3%Number of Params: 1400M

Rank counts only results with a code link.