Datasets:
ArXiv:
License:
File size: 10,826 Bytes
a3341ee |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 |
#!/usr/bin/python3
"""
Evaluation script for the Multi-Camera People Tracking track in the AI City Challenge, Track 1 in 2024.
# Environment setup:
[Optional] conda create -n aicity24-track1 python=3.10
[Optional] conda activate aicity24-track1
pip3 install pandas
pip3 install matplotlib
pip3 install scipy
# Usage: Set number of cores based on your cpu core count.
python3 aicityeval-track1.py --prediction_file ./sample_file/pred.txt --ground_truth_file ./sample_file/ground_truth_test_full.txt --num_cores 16 --scene_2_camera_id_file ./sample_file/scene_name_2_cam_id_full.json
python3 aicityeval-track1.py --prediction_file ./sample_file/pred.txt --ground_truth_file ./sample_file/ground_truth_test_half.txt --num_cores 16 --scene_2_camera_id_file ./sample_file/scene_name_2_cam_id_half.json
"""
import os
import sys
import time
import tempfile
import trackeval
import pandas as pd
import numpy as np
from argparse import ArgumentParser, ArgumentTypeError
from typing import List
from utils.io_utils import load_csv_to_dataframe_from_file, write_dataframe_to_csv_to_file, make_seq_maps_file, make_seq_ini_file, make_dir, check_file_size, get_scene_to_camera_id_dict
def get_unique_entry_per_scene(dataframe, scene_name, scene_2_camera_id):
camera_ids = scene_2_camera_id[scene_name]
filtered_df = dataframe[dataframe["CameraId"].isin(camera_ids)]
unique_entries_df = filtered_df.drop_duplicates(subset=["FrameId", "Id"])
return unique_entries_df
def check_positive(value):
int_value = int(value)
if int_value <= 0:
raise ArgumentTypeError(f"{value} is an invalid num of cores")
return int_value
def computes_mot_metrics(prediction_file_path: str, ground_truth_path: str, output_dir: str, num_cores: int, scene_2_cam_id_file: str) -> None:
check_file_size(prediction_file_path)
# Create a temp directory if output_dir is not specified
is_temp_dir = False
if output_dir is None:
temp_dir = tempfile.TemporaryDirectory()
is_temp_dir = True
output_dir = temp_dir.name
print(f"Temp files will be created here: {output_dir}")
# Create a scene 2 camera_id dict
scene_2_camera_id = get_scene_to_camera_id_dict(scene_2_cam_id_file)
camera_ids = {camera_id for camera_ids in scene_2_camera_id.values() for camera_id in camera_ids}
# Load ground truth and prediction files in dataframe
column_names = ["CameraId", "Id", "FrameId", "X", "Y", "Width", "Height", "Xworld", "Yworld"]
mot_pred_dataframe = load_csv_to_dataframe_from_file(prediction_file_path, column_names, camera_ids)
ground_truth_dataframe = load_csv_to_dataframe_from_file(ground_truth_path, column_names, camera_ids)
# Create evaluater configs for trackeval lib
default_eval_config = trackeval.eval.Evaluator.get_default_eval_config()
default_eval_config["PRINT_CONFIG"] = False
default_eval_config["USE_PARALLEL"] = True
default_eval_config["LOG_ON_ERROR"] = None
default_eval_config["NUM_PARALLEL_CORES"] = num_cores
# Create dataset configs for trackeval lib
default_dataset_config = trackeval.datasets.MotChallenge3DLocation.get_default_dataset_config()
default_dataset_config["DO_PREPROC"] = False
default_dataset_config["SPLIT_TO_EVAL"] = "all"
default_dataset_config["GT_FOLDER"] = os.path.join(output_dir, "evaluation", "gt")
default_dataset_config["TRACKERS_FOLDER"] = os.path.join(output_dir, "evaluation", "scores")
default_dataset_config["PRINT_CONFIG"] = False
# Make output directory for storing results
make_dir(default_dataset_config["GT_FOLDER"])
make_dir(default_dataset_config["TRACKERS_FOLDER"])
# Create sequence maps file for evaluation
seq_maps_file = os.path.join(default_dataset_config["GT_FOLDER"], "seqmaps")
make_seq_maps_file(seq_maps_file, scene_2_camera_id.keys(), default_dataset_config["BENCHMARK"], default_dataset_config["SPLIT_TO_EVAL"])
# Set the metrics to obtain
default_metrics_config = {"METRICS": ["HOTA"], "THRESHOLD": 0.5}
default_metrics_config["PRINT_CONFIG"] = False
config = {**default_eval_config, **default_dataset_config, **default_metrics_config} # Merge default configs
eval_config = {k: v for k, v in config.items() if k in default_eval_config.keys()}
dataset_config = {k: v for k, v in config.items() if k in default_dataset_config.keys()}
metrics_config = {k: v for k, v in config.items() if k in default_metrics_config.keys()}
# Create prediction and ground truth list
for scene_name in scene_2_camera_id.keys():
# Convert ground truth multi-camera dataframe to single camera in MOT format
ground_truth_dataframe_per_scene = get_unique_entry_per_scene(ground_truth_dataframe, scene_name, scene_2_camera_id)
ground_truth_dataframe_per_scene = ground_truth_dataframe_per_scene[["FrameId", "Id", "X", "Y", "Width", "Height", "Xworld", "Yworld"]]
ground_truth_dataframe_per_scene = ground_truth_dataframe_per_scene.sort_values(by="FrameId")
# Make ground truth frame-ids 1-based
ground_truth_dataframe_per_scene["FrameId"] += 1
# Set other defaults
ground_truth_dataframe_per_scene["Conf"] = 1
ground_truth_dataframe_per_scene["Zworld"] = -1
ground_truth_dataframe_per_scene = ground_truth_dataframe_per_scene[["FrameId", "Id", "X", "Y", "Width", "Height", "Conf", "Xworld", "Yworld", "Zworld"]]
# Remove logs for negative frame ids
ground_truth_dataframe_per_scene = ground_truth_dataframe_per_scene[ground_truth_dataframe_per_scene["FrameId"] >= 1]
# Save single camera ground truth in MOT format as CSV
mot_version = default_dataset_config["BENCHMARK"] + "-" + default_dataset_config["SPLIT_TO_EVAL"]
gt_dir = os.path.join(default_dataset_config["GT_FOLDER"], mot_version)
dir_name = os.path.join(gt_dir, str(scene_name))
gt_file_dir = os.path.join(gt_dir, str(scene_name), "gt")
gt_file_name = os.path.join(gt_file_dir, "gt.txt")
make_dir(gt_file_dir)
write_dataframe_to_csv_to_file(gt_file_name, ground_truth_dataframe_per_scene)
# Convert predicted multi-camera dataframe to MOT format
mot_pred_dataframe_per_scene = get_unique_entry_per_scene(mot_pred_dataframe, scene_name, scene_2_camera_id)
mot_pred_dataframe_per_scene = mot_pred_dataframe_per_scene[["FrameId", "Id", "X", "Y", "Width", "Height", "Xworld", "Yworld"]]
mot_pred_dataframe_per_scene = mot_pred_dataframe_per_scene.sort_values(by="FrameId")
# Make MOT prediction frame-ids 1-based
mot_pred_dataframe_per_scene["FrameId"] += 1
# Remove logs for negative frame ids
mot_pred_dataframe_per_scene = mot_pred_dataframe_per_scene[mot_pred_dataframe_per_scene["FrameId"] >= 1]
# Set other defaults
mot_pred_dataframe_per_scene["Conf"] = 1
mot_pred_dataframe_per_scene["Zworld"] = -1
mot_pred_dataframe_per_scene = mot_pred_dataframe_per_scene[["FrameId", "Id", "X", "Y", "Width", "Height", "Conf", "Xworld", "Yworld", "Zworld"]]
# Save single camera prediction in MOT format as CSV
mot_file_dir = os.path.join(default_dataset_config["TRACKERS_FOLDER"], mot_version, "data", "data")
make_dir(mot_file_dir)
tracker_file_name = str(scene_name) + ".txt"
mot_file_name = os.path.join(mot_file_dir, tracker_file_name)
write_dataframe_to_csv_to_file(mot_file_name, mot_pred_dataframe_per_scene)
# Make sequence ini file for trackeval library
if np.isnan(mot_pred_dataframe_per_scene["FrameId"].max()):
last_frame_id = ground_truth_dataframe_per_scene["FrameId"].max()
elif np.isnan(ground_truth_dataframe_per_scene["FrameId"].max()):
last_frame_id = mot_pred_dataframe_per_scene["FrameId"].max()
else:
last_frame_id = max(mot_pred_dataframe_per_scene["FrameId"].max(), ground_truth_dataframe_per_scene["FrameId"].max())
make_seq_ini_file(dir_name, scene=str(scene_name), seq_length=last_frame_id)
# Evaluate ground truth & prediction to get all exhaustive metrics
evaluator = trackeval.eval.Evaluator(eval_config)
dataset_list = [trackeval.datasets.MotChallenge3DLocation(dataset_config)]
temp_metrics_list = [trackeval.metrics.HOTA]
metrics_list = []
for metric in temp_metrics_list:
if metric.get_name() in metrics_config["METRICS"]:
metrics_list.append(metric(metrics_config))
results = evaluator.evaluate(dataset_list, metrics_list)
if is_temp_dir:
temp_dir.cleanup()
return results
def evaluate(prediction_file: str, ground_truth_file: str, output_dir: str, num_cores: int, scene_2_camera_id_file: str) -> None:
# Collect the result
sequence_result = computes_mot_metrics(prediction_file, ground_truth_file, output_dir, num_cores, scene_2_camera_id_file)
# Compute average
final_result = dict()
HOTA_scores = []
DetA_scores = []
AssA_scores = []
LocA_scores = []
for scene_name, result in sequence_result[0]["MotChallenge3DLocation"]["data"].items():
if scene_name == "COMBINED_SEQ":
continue
result = result["pedestrian"]["HOTA"]
HOTA_scores.append(np.mean(result["HOTA"]))
DetA_scores.append(np.mean(result["DetA"]))
AssA_scores.append(np.mean(result["AssA"]))
LocA_scores.append(np.mean(result["LocA"]))
final_result["FINAL"] = dict()
final_result["FINAL"]["HOTA"] = np.mean(np.array(HOTA_scores))
final_result["FINAL"]["DetA"] = np.mean(np.array(DetA_scores))
final_result["FINAL"]["AssA"] = np.mean(np.array(AssA_scores))
final_result["FINAL"]["LocA"] = np.mean(np.array(LocA_scores))
return final_result
if __name__ == '__main__':
start = time.time()
# Parse arguments
parser = ArgumentParser()
parser.add_argument('--prediction_file', required=True)
parser.add_argument('--ground_truth_file', required=True)
parser.add_argument('--output_dir')
parser.add_argument('--num_cores', type=check_positive, default=1)
parser.add_argument('--scene_2_camera_id_file', required=True)
args = parser.parse_args()
# Run evaluation
final_result = evaluate(args.prediction_file, args.ground_truth_file, args.output_dir, args.num_cores, args.scene_2_camera_id_file)
end = time.time()
print(f"Total runtime: {end-start} seconds.")
print(f"HOTA: {float(final_result['FINAL']['HOTA'] * 100):.4f}%")
print(f"DetA: {float(final_result['FINAL']['DetA'] * 100):.4f}%")
print(f"AssA: {float(final_result['FINAL']['AssA'] * 100):.4f}%")
print(f"LocA: {float(final_result['FINAL']['LocA'] * 100):.4f}%")
|