CVMI-Lab / CoDet

(NeurIPS2023) CoDet: Co-Occurrence Guided Region-Word Alignment for Open-Vocabulary Object Detection
110 stars 7 forks source link

lvis_v1_train_norare_cat_info.json #6

Closed junnnxie closed 12 months ago

junnnxie commented 12 months ago

Hello, when I tried to use config CoDet_OVLVIS_SwinB_4x_ft4x.yaml to train, a prompt appeared that the file lvis_v1_train_norare_cat_info.json was missing. Can you tell me how to obtain this file? Thank you very much

machuofan commented 12 months ago

Hi, you need to run

python tools/remove_lvis_rare.py --ann datasets/lvis/lvis_v1_train.json

Please refer to DATA.md for details.

junnnxie commented 12 months ago

Thank you for your reply. I generated lvis_v1_train_norare.json according to DATA.md, but the configuration file requires lvis_v1_train_norare_cat_info.json. Are these two the same file, or does the later file need to be generated separately?

machuofan commented 12 months ago

Sorry. My mistake. You can get lvis_v1_train_norare_cat_info.json by running

python tools/get_lvis_cat_info.py --ann datasets/lvis/lvis_v1_train_norare.json --add_freq
junnnxie commented 12 months ago

Thx a lot!