Evaluate by dataset velocity (#89)

Author: francodipietro

evaluation/velocity_study.py

@@ -0,0 +1,151 @@
+import argparse
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+import seaborn as sns
+from tqdm import tqdm
+from pathlib import Path
+from scipy import interpolate
+from datasets.utils import constants as ds_constants, parsing as ds_parsing
+
+        
+        
+def avg_vel(ds_name, timestamps, joints_gt):
+    
+    assert 1 <= joints_gt.shape[2] <= 3, 'coordinates must be either 2D or 3D'
+    # iterate on each joint
+    data = np.zeros((len(timestamps), len(ds_constants.HPECoreSkeleton.KEYPOINTS_MAP) * 2))
+    keys = []
+    for joint_key, joint_ind in ds_constants.HPECoreSkeleton.KEYPOINTS_MAP.items():
+        data[:,joint_ind*2] = joints_gt[:, joint_ind, 0]
+        data[:,joint_ind*2+1] = joints_gt[:, joint_ind, 1]
+        keys.append(joint_key + ' x')
+        keys.append(joint_key + ' y')
+
+    df = pd.DataFrame(abs(data), columns=keys)
+    # print(df.describe())
+    avg75 = np.mean(df.describe().loc['75%'])
+    # print("\n75% = " + str(avg75), flush=True)
+    avgMax = np.mean(df.describe().loc['max'])
+    # print("Max = " + str(avgMax), flush=True)
+    avgMean = np.mean(df.describe().loc['mean'])
+    # print("Mean = " + str(avgMean), flush=True)
+    avgStd = np.mean(df.describe().loc['std'])
+    # print("Std = " + str(avgStd), flush=True)    
+    
+    if(avgMax >= 250 and avgMean >=57):
+        print("FAST")
+    elif(avgMax < 100 and avgMean < 18):
+        print("SLOW")
+   
+    return avg75, avgMax, avgMean, avgStd
+            
+
+def main(args):
+    
+    plt.close('all')
+
+    # output_folder_path = Path(args.output_folder)
+    # output_folder_path = output_folder_path.resolve()
+    # output_folder_path.mkdir(parents=True, exist_ok=True)
+    results = dict()
+    results['datasets'] = dict()
+    results['global'] = dict()
+    list75 = []
+    listMax = []
+    listMean = []
+    listStd = []
+    
+    # import GT from yarp
+    datasets_path = Path(args.datasets_path)
+    yarp_file_paths = list(datasets_path.glob('**/data.log'))
+
+    for yarp_path in tqdm(yarp_file_paths, desc ="Progress: "):
+    # for yarp_path in track(yarp_file_paths):
+
+        if 'skeleton' not in yarp_path.parent.name:
+            continue
+        dataset_name = yarp_path.parent.parent.name
+        print('\x1b[1;33;20m' + "Procesing " + str(dataset_name)  + '\x1b[0m')
+        data = ds_parsing.import_yarp_skeleton_data(yarp_path)
+        ts_gt = np.concatenate(([.0], data['ts'], [data['ts'][-1] + 1]))
+    
+        # interpolate ground truth joints so that they can be compared with the high frequency predictions
+        for k_map in ds_constants.HPECoreSkeleton.KEYPOINTS_MAP.items():
+            x_interpolation = interpolate.interp1d(ts_gt, np.concatenate(([data[k_map[0]][0, 0]], data[k_map[0]][:, 0], [data[k_map[0]][-1, 0]])))
+            y_interpolation = interpolate.interp1d(ts_gt, np.concatenate(([data[k_map[0]][0, 1]], data[k_map[0]][:, 1], [data[k_map[0]][-1, 1]])))
+            data[k_map[0]] = dict()
+            data[k_map[0]]['x'] = x_interpolation
+            data[k_map[0]]['y'] = y_interpolation
+        skeletons_gt = np.zeros((len(ts_gt), len(ds_constants.HPECoreSkeleton.KEYPOINTS_MAP), 2))
+        for k_map in ds_constants.HPECoreSkeleton.KEYPOINTS_MAP.items():
+            skeletons_gt[:, k_map[1], 0] = data[k_map[0]]['x'](ts_gt)
+            skeletons_gt[:, k_map[1], 1] = data[k_map[0]]['y'](ts_gt)
+  
+        # resample GT at 1 KHz
+        fHz = 1000
+        tGT1K = np.arange(ts_gt[0], ts_gt[-1], 1/fHz)
+        skeletons_gt1K = np.zeros((len(tGT1K), len(ds_constants.HPECoreSkeleton.KEYPOINTS_MAP), 2))
+        for k_map in ds_constants.HPECoreSkeleton.KEYPOINTS_MAP.items():
+            skeletons_gt1K[:, k_map[1], 0] = data[k_map[0]]['x'](tGT1K)
+            skeletons_gt1K[:, k_map[1], 1] = data[k_map[0]]['y'](tGT1K)
+        # differenciate GT poses
+        vel_gt1K_noF = np.gradient(skeletons_gt1K, axis=0)*fHz
+        # filter velocity
+        from scipy.signal import savgol_filter
+        vel_gt1K = savgol_filter(vel_gt1K_noF, 901, 3, axis=0) 
+        
+        # calculate statistics for each dataset
+        value75, valueMax, valueMean, valueStd = avg_vel(str(dataset_name), tGT1K, vel_gt1K)
+        # add to global list
+        list75.append(value75)
+        listMax.append(valueMax)
+        listMean.append(valueMean)
+        listStd.append(valueStd)
+    
+    dataList = list(zip(list75, listMax, listMean, listStd))
+    df = pd.DataFrame(dataList, columns=['75%', 'Max', 'Mean', 'Std'])
+    # print(df.describe())
+    
+    # define clusters for mean velocity
+    df['mean_group'] = pd.cut(df['Mean'], bins=range(5, 71, 13))
+    # define clusters for max velocity
+    df['max_group'] = pd.cut(df['Max'], bins=range(50, 301, 50))
+    
+    # plots
+    my_dpi = 96
+    fig1 = plt.figure(figsize=(2048/my_dpi, 900/my_dpi), dpi=my_dpi)
+    ax1 = sns.boxplot(x="mean_group", y="Mean", data=df)
+    ax1.set_xlabel("Speed [px/s]", fontsize=24)
+    ax1.set_ylabel("Speed [px/s]", fontsize=24)
+    fig1.suptitle('Mean velocity distribution', fontsize=32, y=0.92)
+    plt.xticks(fontsize=22, rotation=0)
+    
+    fig2 = plt.figure(figsize=(2048/my_dpi, 900/my_dpi), dpi=my_dpi)
+    ax2 = sns.boxplot(x="max_group", y="Max", data=df)
+    ax2.set_xlabel("Speed [px/s]", fontsize=24)
+    ax2.set_ylabel("Speed [px/s]", fontsize=24)
+    fig2.suptitle('Max velocity distribution', fontsize=32, y=0.92)
+    plt.xticks(fontsize=22, rotation=0)
+    
+    # print statistics
+    print(df)
+    print("Mean values range:")
+    print("[", np.min(df['Mean']), ", ", np.max(df['Mean']), "]")
+    print("Max values range:")
+    print("[", np.min(df['Max']), ", ", np.max(df['Max']), "]")
+   
+    plt.show()
+        
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='...')
+    parser.add_argument('-d', '--datasets_path', help='Path to the folders containing data saved in Yarp format', required=True)
+    parser.add_argument('-o', '--output_folder', help='Path to the folder where velocity study results will be saved', required=False)
+
+    args, unknown = parser.parse_known_args()
+    if(unknown):
+        print('\x1b[1;31;20m' + 'Unknown argument/s: ' + ' '.join(unknown) + '\x1b[0m')
+
+    main(args)