CVAT XML To TXT - It only works for interpolation mode

cvat_xml_to_txt.py

import os
import argparse
from math import floor
import xml.etree.ElementTree as ET

# Parse Arguments
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument("--input", type=str, help="Input XML file", required=True,
                        dest="input_xml")
arg_parser.add_argument("--output", type=str, help="Output .txt filename", required=True,
                        dest="output_txt")
arg_parser.add_argument("--frame_dir", type=str, help="Directory where frames exists", required=True,
                        dest="frame_dir")
arg_parser.add_argument("--frame_format_string", type=str, default="frame%06d.jpg")
args = arg_parser.parse_args()


xml_obj = ET.parse(args.input_xml)
root = xml_obj.getroot()
training_image_path = "frames"

tracks = [child for child in root if child.tag == "track"]

with open(args.output_txt, "w") as output_file:
    for track in tracks:
        for box in track:
            obj = box.attrib
            
            if obj['outside'] == "0":
                filename = args.frame_format_string % int(obj['frame'])
                filepath = os.path.join(args.frame_dir, filename)
                
                formated_obj = [filepath,
                                floor(float(obj['xtl'])),
                                floor(float(obj['ytl'])),
                                floor(float(obj['xbr'])),
                                floor(float(obj['ybr'])),
                                track.attrib['label'].lower()]
                
                formated_obj_str = list(map(str, formated_obj))
                
                str_to_write = ",".join(formated_obj_str) + "\n"
                output_file.write(str_to_write)