search

thunlp / PEVL

Source code for EMNLP 2022 paper “PEVL: Position-enhanced Pre-training and Prompt Tuning for Vision-language Models”
MIT License
47 stars 5 forks source link

token position #1

Closed chenjianqiang199526 closed 2 years ago

chenjianqiang199526 commented 2 years ago

Thank author for prviding clear code.I wonder how to get the tokens positive and tokens negative?

qyc-98 commented 2 years ago

our data mainly came from MDETR except VCR，and this may help you.