extract_triplet.py

from bschecker.extractor import GPT4Extractor
import json

from tqdm import tqdm

## Import the file with an LVLM's answers
## we suggest you directly store the model's answer as an item after each "question" item in the "tri-he.json" file.
## This helps you better align your results with the following codes.
with open('xxx.json', 'r') as file:
    answers = json.load(file)

## ids of all the 300 images
questions_ids = ['2374892', '2397582', '2387230', '2383415', '2380479', '2402409', '2391748', '2404666', '2315880', '2331017', '2347569', '2319825', '2325480', '2407148',
                    '2402232', '2405227', '2359433', '2396963', '2333677', '2395042', '2363853', '2362053', '2386393', '2410230', '2337344', '2317575', '2367982', '1592403',
                    '2367184', '2395567', '2414858', '2406059', '2371404', '2379186', '2399534', '286053', '2380125', '2325937', '2395794', '2385456', '2317148', '2412131', '2335787',
                    '2331948', '2391000', '2397448', '2415899', '2341361', '2358708', '2323530', '2379961', '713360', '2402763', '2393510', '2320543', '2412046', '2380680', '2396106',
                    '2347091', '2409544', '2387777', '2337304', '2376529', '2353439', '3455', '2396483', '2390229', '2367328', '2406011', '2346625', '2365078', '2368472', '2393281',
                    '2326674', '2375491', '2384779', '2384092', '2366958', '2406666', '2410898', '2407164', '2355023', '2318090', '2386254', '2386334', '2404613', '2383991', '2341840',
                    '2365562', '2388962', '2391950', '2319558', '2389203', '2389205', '2409543', '2410392', '2319034', '2411516', '2355548', '2383328', '1592175', '2404536', '2411375',
                    '2412542', '2364025', '2320507', '2394753', '2376660', '2393164', '2380977', '2380375', '2362253', '314', '2393775', '2325932', '2362477', '2335178', '2328879',
                    '2341512', '2403003', '2381595', '2376620', '2337945', '2386867', '2401166', '2363959', '2404602', '2408596', '2355237', '2316628', '2329334', '2379906', '2397488',
                    '2359140', '2371945', '2411488', '2350736', '2404849', '2331876', '2401506', '2392665', '2406998', '2387508', '2391364', '2343796', '2415157', '2400560', '2385013',
                    '2342134', '2390108', '2348072', '2323054', '2397341', '2366876', '2390911', '2391205', '2390599', '2333858', '2861', '2391835', '2342333', '2323323', '2382847', '2405830',
                    '2347819', '2384812', '2368361', '2414129', '2272', '2336635', '2400789', '2405555', '2407415', '2323268', '2376766', '1591820', '2347171', '2396799', '2325882', '2320615',
                    '2399206', '2341698', '2364886', '2384012', '2369640', '2326613', '2403060', '2388598', '2355736', '2407347', '2327560', '2402334', '2357353', '2356008', '2387752',
                    '2406309', '2341341', '2323243', '2342451', '2372614', '2401249', '2378167', '2394962', '2352009', '2356494', '713531', '2346104', '2377608', '2393430', '2359715',
                    '2371761', '2374803', '2383613', '2383614', '2414693', '2353016', '2320407', '2406323', '2343415', '2412204', '2407451', '2383582', '2402165', '2316099', '70',
                    '2353731', '2322017', '2318853', '2343509', '2370948', '2380022', '2354914', '2356041', '2399053', '2348163', '1592683', '2352342', '2317627', '2386682', '2385133',
                    '2326672', '2403546', '2389793', '2320676', '2378619', '2373168', '2362327', '2400175', '2390700', '2385788', '1159264', '2398554', '2351181', '2375868', '2320339',
                    '2346506', '2339309', '2346094', '1592103', '1159845', '2400618', '2366184', '2400603', '2327311', '2371558', '2368487', '2406620', '2251', '2374729', '2340730', '2369531',
                    '2342152', '2389148', '2397079', '2393941', '2397910', '2366510', '2368283', '2375806', '2343787', '2405896', '2409391', '2378861', '3898', '2414630', '2387206', '2362421',
                    '2344585', '2319086', '2398388', '2367256', '2346063', '2331048', '2360708', '2401215', '2412723', '2315718', '2387283', '2322244', '2409876']

for index in tqdm(questions_ids):

    for question in answers[index]['instance']:

        answer = question["the key of the item that stores the LVLM's response"]
        response = (answer)
        extractor = GPT4Extractor()
        triplets = extractor.extract(response)
        question['xxx_triplets'] = triplets

    with open('xxx_with_triplets.json', 'w') as file:
        json.dump(answers, file)