TokenLearner: What Can 8 Learned Tokens Do for Images and Videos?

Open paper
Benchmark Model Rank Results
action-classification-on-avidTokenLearner#1Accuracy: 53.8
action-classification-on-charadesTokenLearner#1MAP: 66.3
action-classification-on-kinetics-400TokenLearner 16at18 (L/10)#46Acc@1: 85.4
action-classification-on-kinetics-600TokenLearner 16at18 w. Fuser (L/10)#21Top-1 Accuracy: 86.3Top-5 Accuracy: 97.0
image-classification-on-imagenetTokenLearner L/8 (24+11)#27Top 1 Accuracy: 88.87%Number of params: 460M
image-classification-on-imagenet16-TokenLearner B/16 (21)#107Top 1 Accuracy: 87.07%
image-classification-on-imagenet-realTokenLearner L/8 (24+11)#6Accuracy: 91.05%Params: 460M

Rank counts only results with a code link.