search

aurooj / WSG-VQA-VLTransformers

Weakly Supervised Grounding for VQA in Vision-Language Transformers
MIT License
16 stars 2 forks source link

Issues about the pre-training code: "img_id2idx_train.json" couldn't be obtained #2

Closed leyi-123 closed 1 year ago

leyi-123 commented 1 year ago

Hello, when I reproduced the pre-training code and ran the "bash run/2stage_fulldata_no_init_16_caps.bash" command, I found that the data file "img_id2idx_train.json" is needed, which is in line 91 in src/utils.py. I wonder how this file was obtained. Thanks for your reply.

zxzhou9 commented 1 year ago

I meet the same problem. Have you ever solved it?

aurooj commented 1 year ago

Hello, I have uploaded the required files img_id2idx_train.json and img_id2idx_valid.json under data_files/ directory on github repo. Thanks for your patience.

zxzhou9 commented 1 year ago

Thank you so much! Really appreciate that.

zxzhou9 commented 1 year ago

Sorry to bother you again, I find the gqa_sqatial_merged_info.json can not be found too. Could you upload it later?

aurooj commented 1 year ago

Hi! You can download and preprocess the GQA dataset from their official github repo. Follow instructions for Pre-processing section will yield you gqa_sqatial_merged_info.json file.

I hope it helps.