dandelin / ViLT

Code for the ICML 2021 (long talk) paper: "ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision"
Apache License 2.0
1.36k stars 209 forks source link

COCO split for pre-training #49

Open sanjayss34 opened 2 years ago

sanjayss34 commented 2 years ago

Hi @dandelin , thanks for this great repo and work! Could you please say what COCO split was used for pre-training? (was it 2014, 2017, Karpathy, or something else?) Thanks!