clip4clip_prep.py

import json
from preprocess_pkg.clip4clip import PiaTemplateJson2UniversalFormat, MakePiaTemplateJsonFromClassifiedFolders, MainUtils, CLIP4ClipPrep, SyntheticCaptionGeneration
import os

if __name__ == "__main__":
    MU = MainUtils()
    skip_folder = []

    # NAS_PATH = "/home/piawsa6000/nas192"
    NAS_PATH = "/mnt/nas192"

    input_path_1 = os.path.join(NAS_PATH, "Research_materials/PIA_clip_dataset/CLIP4Clip_format/PIA_clip_outdoor_v2/original_train_set_processed")
    input_path_2 = os.path.join(NAS_PATH, "Research_materials/PIA_clip_dataset/CLIP4Clip_format/PIA_clip_outdoor_v2/original_val_test_set_processed")
    pia_template_json_output_path = "./"
    universal_format_output_path = "pia_template_universal_format_combined.json"


    folder2classmapping = {
        "falldown": "falldown",
        "fire": "fire",
        "normal": "normal",
        "violence": "violence",
        "smoking": "smoking"
    }

    json_out_path = MakePiaTemplateJsonFromClassifiedFolders(input_path_1,
                                        pia_template_json_output_path,
                                        skip_folder,
                                        folder2classmapping,
                                        train_split_param = 0.05,
                                        test_split_param = 0.5,
                                        copy_video = True,
                                        multi_process_copy = True).run()

    universal_format_1 = PiaTemplateJson2UniversalFormat(json_out_path).get_universal_format()

    json_out_path = MakePiaTemplateJsonFromClassifiedFolders(input_path_2,
                                        pia_template_json_output_path,
                                        skip_folder,
                                        folder2classmapping,
                                        train_split_param = 0.05,
                                        test_split_param = 0.5,
                                        copy_video = True,
                                        multi_process_copy = True).run()

    universal_format_2 = PiaTemplateJson2UniversalFormat(json_out_path).get_universal_format()

    combined_universal_format = MU.combine_two_dataset(universal_format_1, universal_format_2, "train", "val_test")
    print(combined_universal_format)

    json_out_path = MU._save_as_json(combined_universal_format, 
                     pia_template_json_output_path,
                     json_name = universal_format_output_path)


    C4C = CLIP4ClipPrep(json_out_path)
    # C4C.make_train_val_test()
    C4C.make_class_wise_cfg()
    C4C._video_caption_dict_process()
    C4C.make_formated_caption()
    # C4C.checker_train_val_test()

    synthetic_template_placeholder_caption = "./synthethic_caption_template_placeholder.json"

    generator = SyntheticCaptionGeneration(synthetic_template_placeholder_caption)


    C4C = CLIP4ClipPrep(None, output_name = "synthethic_caption_CLIP4CLIP")


    for video_id, data in combined_universal_format.items():
        caption = generator.make_caption(data["category"].lower())
        C4C._video_synthetic_caption_dict_process(video_id, caption)
    C4C.make_formated_caption()