lisa_data/sample.py

import json
import random

data_list = [
    'ade_seg.json',
    'cocostuff_seg.json',
    'mapillary_seg.json',
    'paco_lvis_seg.json',
    'pascal_part_seg.json',
    # 'llava_instruct_150k.json'
]
final_data = []

ref_data = json.load(open('ref_all.json'))

for d in data_list:
    data = json.load(open(d))
    print(d, ' ', len(data))
    final_data+=data

random.shuffle(final_data)

output_data = final_data[:200000]
output_data+=ref_data

print(len(output_data))
with open('sample_seg_300k.json', 'w') as f:
    f.write(json.dumps(output_data, indent=4))