UCSB-NLP-Chang/Diffusion-SpaceTime-Attn

Where does mscoco.txt under the datasets floder come from coco2017‘val

Closed this issue · 2 comments

Maybeetw commented

Hi there,

Thank you for your amazing work.I would like to ask about the 500 data in mscoco.txt under the datasets folder. These 500 captions are from the raw coco data stuff_ Val.json or caption_ Val.json? It seems like I can't find these 500 texts here, or are these 500 texts from other files in Coco

wuqiuche commented

Hi Maybeetw,

Thank you very much for your interests! These captions are from this link in original COCO website: 2017 Train/Val annotations [241MB].

After downloading, you can find the captions in "captions_train2017.json". For example:

import json
with open("captions_train2017.json", "r") as f:
  contents = json.load(f)
caption_0 = contents['annotations'][0]

The result should be "A bicycle replica with a clock as the front wheel."

Maybeetw commented

Thank you very much for your reply, as the five hundred data were for testing, leading me to think they were from coco_cpation_val or stuff_val,very good job, thanks again for your answer.

  • ❤️1