Add DuckTrack as initial annotation tool; Initial multimodal test

annotation/experiments/stopwatch/stopwatch.py

@@ -0,0 +1,110 @@
+import glob
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+import scipy.stats as stats
+import seaborn as sns
+
+# use this: https://www.estopwatch.net/
+
+def read_file(file_path):
+    df = pd.read_csv(file_path)
+    df['Elapsed time'] = pd.to_datetime(df['Elapsed time'], errors='coerce')
+    return df
+
+
+def analyze_new_error(run_df, groundtruth_df):
+    cumulative_errors = run_df['Elapsed time'] - groundtruth_df['Elapsed time']
+    cumulative_errors_in_seconds = cumulative_errors.dt.total_seconds()
+
+    new_errors_in_seconds = cumulative_errors_in_seconds.diff().fillna(cumulative_errors_in_seconds[0])
+    new_error_points = new_errors_in_seconds[new_errors_in_seconds != 0].index.tolist()
+
+    return new_errors_in_seconds[new_error_points]
+
+def calculate_statistics(errors):
+    if len(errors) == 0:
+        return {
+            'mean_error': 0,
+            'median_error': 0,
+            'stddev_error': 0,
+            'rmse_error': 0,
+            'confidence_interval': (0, 0),
+            'error_frequency': 0
+        }
+
+    mean_error = np.mean(errors)
+    median_error = np.median(errors)
+    stddev_error = np.std(errors)
+    rmse_error = np.sqrt(np.mean(np.square(errors)))
+    
+    ci_low, ci_high = stats.t.interval(
+        confidence=0.95,
+        df=len(errors) - 1,
+        loc=mean_error,
+        scale=stats.sem(errors) if len(errors) > 1 else 0
+    )
+
+    return {
+        'mean_error': mean_error,
+        'median_error': median_error,
+        'stddev_error': stddev_error,
+        'rmse_error': rmse_error,
+        'confidence_interval': (ci_low, ci_high),
+    }
+
+
+def main():
+    groundtruth_file = 'groundtruth.csv'
+    run_files = glob.glob('runs/*.csv')
+
+    groundtruth_df = read_file(groundtruth_file)
+    run_dfs = {f'run{i+1}': read_file(file) for i, file in enumerate(run_files)}
+
+    total_errors = []
+    total_points = 0
+    all_errors = []
+
+    for run, df in run_dfs.items():
+        errors = analyze_new_error(df, groundtruth_df)
+        total_errors.extend(errors)
+        all_errors.extend(errors)
+        total_points += len(df)
+        
+        results = calculate_statistics(errors)
+        error_frequency = len(errors) / len(df)
+        
+        print(f"Results for {run}:")
+        print(f"Mean New Error: {results['mean_error']:.5f} seconds")
+        print(f"Median New Error: {results['median_error']:.5f} seconds")
+        print(f"Standard Deviation of New Error: {results['stddev_error']:.5f} seconds")
+        print(f"RMSE of New Error: {results['rmse_error']:.5f} seconds")
+        print(f"95% Confidence Interval of New Error: ({results['confidence_interval'][0]:.5f}, {results['confidence_interval'][1]:.5f}) seconds")
+        print(f"New Error Frequency: {error_frequency*100:.5f} %")
+        print('-----------------------------------------')
+
+    total_results = calculate_statistics(total_errors)
+    total_error_frequency = len(total_errors) / total_points
+    
+    print("Total Statistics:")
+    print(f"Mean New Error: {total_results['mean_error']:.5f} seconds")
+    print(f"Median New Error: {total_results['median_error']:.5f} seconds")
+    print(f"Standard Deviation of New Error: {total_results['stddev_error']:.5f} seconds")
+    print(f"RMSE of New Error: {total_results['rmse_error']:.5f} seconds")
+    print(f"95% Confidence Interval of New Error: ({total_results['confidence_interval'][0]:.5f}, {total_results['confidence_interval'][1]:.5f}) seconds")
+    print(f"New Error Frequency: {total_error_frequency*100:.5f} %")
+    
+    # do plus minus
+    print(f"New Error: {total_results['mean_error']:.5f} ± {total_results['confidence_interval'][1] - total_results['mean_error']:.5f} seconds")
+    
+    plt.figure(figsize=(10, 5))
+    sns.histplot(all_errors, bins=12, kde=False)
+    plt.title('Distribution of Newly Introduced Errors (macOS)')
+    plt.xlabel('Error Duration (seconds)')
+    plt.ylabel('Frequency')
+    plt.savefig('error_dist', dpi=300)
+    plt.show()
+
+if __name__ == "__main__":
+    main()