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

what is the meaning of "split" in /vilt/utils/write_conceptual_caption.py #54

Open campper opened 2 years ago

campper commented 2 years ago

when do pretrain-training,it raise the exception:"Error while read file idx 51226 in conceptual_caption_val_0 -> cannot identify image file <_io.BytesIO object at 0x7fe894cfcb48>" likely. I made the dataset format as "{image:"./data/conceptual-captions/val_images/XXXX.jpg",caption:[word1,word2,....wordn]}" so i was confused that what's problem of my project, if it maybe has some causing relationship with the "split" keyword.