Convolutional Bypasses Are Better Vision Transformer Adapters

Open paper
Benchmark Model Rank Results
monocular-depth-estimation-on-clevr-distanceConvpass ViT-B/16 (VTAB-1k)–Accuracy: 67.9
object-counting-on-clevr-countConvpass ViT-B/16 (VTAB-1k)–Accuracy: 82.3

Rank counts only results with a code link.