import glob
import copy
import json
import pdb
from tqdm import tqdm

dirs = glob.glob('extra_data/SA-V/sav_train/sav_*')

paths = []
for dir_ in dirs:
    paths += glob.glob(dir_ + '/*_object_*.json')

llava_data = []

for sample_id, path in enumerate(tqdm(paths)):
    data = json.load(open(path, 'r'))
    item = {"id": 0, 
            "video": "",
            "conversations": [{"value": "<image> Identify the object according to the following bounding box: \n <OBJ>", "from": "human"}, \
                                {"value": "<OBJ>", "from": "gpt"}], 
            "metadata": {"dataset": "sam2_tracking", "box_info": ""}}
    item["id"] = sample_id
    item["video"] = path[0:44] + '.mp4'
    item["metadata"]["box_info"] = data

    llava_data.append(item)

json.dump(llava_data, open('extra_data/annotation/sam2_tracking_llava_format.json', 'w'), indent=4)