search

microsoft / Oscar

Oscar and VinVL
MIT License
1.04k stars 251 forks source link

Image Captioning on COCO Val Set Mismatches with Images #180

Open tprdk opened 2 years ago

tprdk commented 2 years ago

Hi, I tried to see test results on coco caption14 val dataset and visualize it. I downloaded the coco_caption dataset and pretrained models. like base_xe etc. Some of them creating a sentence like "a a a the the the ,,,," so i tried another models and got some well preprared senteces. Now the problem is, when i try to visualize results, reading image from they key which is getting by dataloader. Image and caption seems unappropriate. Am i doing someting wrong with reading image or img_key value is different from 12 digit img names in coco dataset ?

Thanks for your help.

serdaryildiz commented 2 years ago

same problem. I spent lots of time.