Skip to content

Latest commit

 

History

History
11 lines (6 loc) · 789 Bytes

dataset_prepare.md

File metadata and controls

11 lines (6 loc) · 789 Bytes

Open-Vocabulary COCO

  1. Download the COCO dataset.

  2. Create the annotation jsons for the open-vocabulary setting. We use the scripts provide by OVR-CNN. Then add the object proposals that may cover the novel classes. You can download our pre-generated json file in Google Drive.

  3. Extract the CLIP image features. You can download our pre-generated file in Google Drive, or use the script to extract it by yourself.

Open-Vocabulary LVIS

Under preparation.