Official repository of OFA (ICML 2022). Paper: OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework
Apache License 2.0
2.39k
stars
248
forks
source link
Do the training data for the pretrained OFA include samples from COCO val set? #348
Does the pretrain dataset include COCO detection val, VQA val or COCO captioning val (or part of val)? If yes, it means the pretrained model has seen the images from COCO val (even before finetuning) and there is a data leak problem if I want to evaluate pretrained OFA on the validation set of VQA or COCO Captioning?
Does the pretrain dataset include COCO detection val, VQA val or COCO captioning val (or part of val)? If yes, it means the pretrained model has seen the images from COCO val (even before finetuning) and there is a data leak problem if I want to evaluate pretrained OFA on the validation set of VQA or COCO Captioning?