lisa_data/llava.py

import json

data = json.load(open('/mnt/workspace/workgroup/yuanyq/code/LISA/dataset/llava_dataset/llava_instruct_150k.json'))

for d in data:
    d['image'] = 'coco/train2017/'+d['image']

with open('llava_instruct_150k.json', 'w') as f:
    f.write(json.dumps(data, indent=4))