#!/usr/bin/env python3 import os import json import argparse import math from copy import deepcopy import cv2 import numpy as np import math import sys from prettytable import PrettyTable from colorama import Style, Fore sys.path.insert(1, '../utils') from links import joint_correspondence_human36m, joint_correspondence, joint_correspondence_human36m_mediapipe, links_mpi_inf_3dhp, joints_mpi_inf_3dhp, joint_correspondence_mpi_mediapipe from projection import projectToCamera from draw import drawSquare2D, drawCross2D def main(): # -------------------------------------------------------- # Arguments # -------------------------------------------------------- ap = argparse.ArgumentParser() ap.add_argument("-dataset", "--dataset_name", help="Dataset name. Datasets must be inside hpe/images/. Please use -dataset human36m to optimize Human 3.6M.", type=str, default='mpi') ap.add_argument("-s", "--section", help="Human 3.6M section to optimize. Only works when the dataset is set to human36m", type=str, default='S1') ap.add_argument("-a", "--action", help="Human 3.6M action to optimize. Only works when the dataset is set to human36m", type=str, default='Directions-1') ap.add_argument("-seq", "--sequence", help="MPI sequence optimize. Only works when the dataset is set to human36m", type=str, default='Seq1') ap.add_argument("-si", "--show_images", help="Shows projection images", action="store_true", default=False) ap.add_argument("-ext", "--extended_tables", help="Shows extended tables", action="store_true", default=False) ap.add_argument("-2d_detector", "--2d_detector", help="2D detector used for 2D human pose estimation. Current options: openpose (default), mediapipe", type=str, default='mediapipe') args = vars(ap.parse_args()) # -------------------------------------------------------- # Loading files, and verifications configurations # -------------------------------------------------------- world_frame = 'world' dataset_folder = '../../images/' + args['dataset_name'] + '/' human36m = False mpi = False detector = 'openpose' if args['2d_detector'] == "mediapipe": detector = 'mediapipe' if args['dataset_name'] == "human36m": human36m = True dataset_folder = '../../images/human36m/processed/' + args['section'] + '/' + args['action'] + '/' joint_correspondence = joint_correspondence_human36m if detector == "mediapipe": joint_correspondence = joint_correspondence_human36m_mediapipe elif args['dataset_name'] == "mpi": mpi = True dataset_folder = '../../images/mpi-inf-3dhp/' + args['section'] + '/' + args['sequence'] + '/' links_evaluation = links_mpi_inf_3dhp if detector == "mediapipe": joint_correspondence = joint_correspondence_mpi_mediapipe # check if dataset folder exists: if not os.path.exists(dataset_folder): raise Exception("The dataset folder does not exist!") ground_truth_path = dataset_folder + "ground_truth.txt" if os.path.exists(ground_truth_path): with open(ground_truth_path, "r") as fp: ground_truth = json.load(fp) else: raise Exception("The ground truth file does not exist!") if detector == 'openpose': camera_dict_path = dataset_folder + "cameras_op.json" elif detector == 'mediapipe': camera_dict_path = dataset_folder + "cameras_mp.json" if os.path.exists(camera_dict_path): with open(camera_dict_path, "r") as fp: cameras = json.load(fp) else: raise Exception("The camera dictionary file does not exist!") ground_truth_path = dataset_folder + "ground_truth.txt" if os.path.exists(ground_truth_path): with open(ground_truth_path, "r") as fp: ground_truth = json.load(fp) else: raise Exception("The ground truth file does not exist!") # -------------------------------------------------------- # Calculate projections # -------------------------------------------------------- e = {} for camera_key, camera in cameras.items(): e[camera_key] = {} for frame_key, frame in camera['frames'].items(): window_name = camera_key + '_' + frame_key img = cv2.imread(frame['image_path']) # img = np.uint8(np.array(frame['image'])) e[camera_key][frame_key] = {} for joint_key, joint in joint_correspondence.items(): # print(frame) joint_key_detected = joint['body25'] joint_detected = frame['joints'][joint_key_detected] if not joint_detected['valid']: # skip invalid joints continue e[camera_key][frame_key][joint_key] = {} joint_key_gt = joint['ground_truth'] joint_gt = ground_truth[frame_key][joint_key_gt]['pose'] # get 3d point coordinates pts_in_world = np.ndarray((4, 1), dtype=float) pts_in_world[0][0] = ground_truth[frame_key][joint_key_gt]['pose']['x'] pts_in_world[1][0] = ground_truth[frame_key][joint_key_gt]['pose']['y'] pts_in_world[2][0] = ground_truth[frame_key][joint_key_gt]['pose']['z'] pts_in_world[3][0] = 1 # Transform to the camera's coordinate frame pts_in_sensor = np.dot(camera['extrinsics'], pts_in_world) # Project 3D point to camera pts_in_image, _, _ = projectToCamera(np.array(camera['intrinsics']), np.array(camera['distortion']), camera['width'], camera['height'], pts_in_sensor[0:3, :]) # Compute distance from 2D detection of joint and 3D->2D projection xpix_projected = pts_in_image[0][0] ypix_projected = pts_in_image[1][0] xpix_detected = joint_detected['x'] ypix_detected = joint_detected['y'] color = joint_detected['color'] drawSquare2D(img, xpix_projected, ypix_projected, 10, color=color, thickness=3) drawCross2D(img, xpix_detected, ypix_detected, 10, color=color, thickness=3) # exit(0) x_error = abs(xpix_detected - xpix_projected) y_error = abs(ypix_detected - ypix_projected) rmse = abs(math.dist([xpix_detected, ypix_detected], [xpix_projected, ypix_projected])) e[camera_key][frame_key][joint_key]['x_error'] = x_error e[camera_key][frame_key][joint_key]['y_error'] = y_error e[camera_key][frame_key][joint_key]['rmse'] = rmse if args['show_images']: if detector == 'mediapipe': img_gui= cv2.resize(img, None, fx = 0.5, fy = 0.5) cv2.imshow(window_name,img_gui) else: cv2.imshow(window_name, img) if args['show_images']: cv2.waitKey(0) # print(e) # ------------------------------------------------------------- # Print output table # ------------------------------------------------------------- table_header = ['Camera', 'Frame #', 'Joint', 'RMS (pix)', 'X err (pix)', 'Y err (pix)'] # table = PrettyTable(table_header) if not args['extended_tables']: table = PrettyTable(table_header) for camera_key, camera in cameras.items(): for frame_key, frame in camera['frames'].items(): x_avg = 0 y_avg = 0 rmse_avg = 0 total_joints = 0 for joint_key, joint in joint_correspondence.items(): joint_key_detected = joint['body25'] joint_detected = frame['joints'][joint_key_detected] if not joint_detected['valid']: # skip invalid joints continue x_avg = x_avg + e[camera_key][frame_key][joint_key]['x_error'] y_avg = y_avg + e[camera_key][frame_key][joint_key]['y_error'] rmse_avg = rmse_avg + e[camera_key][frame_key][joint_key]['rmse'] total_joints = total_joints + 1 x_avg = x_avg / total_joints y_avg = y_avg / total_joints rmse_avg = rmse_avg / total_joints avg_row = [camera_key, frame_key, Fore.BLUE + Style.BRIGHT + 'Average' + Fore.BLACK + Style.NORMAL, '%.3f' % rmse_avg, '%.3f' % x_avg, '%.3f' % y_avg] table.add_row(avg_row) print(table) else: for camera_key, camera in cameras.items(): table = PrettyTable(table_header) for frame_key, frame in camera['frames'].items(): x_avg = 0 y_avg = 0 rmse_avg = 0 total_joints = 0 for joint_key, joint in joint_correspondence.items(): joint_key_detected = joint['body25'] joint_detected = frame['joints'][joint_key_detected] if not joint_detected['valid']: # skip invalid joints continue print(e[camera_key][frame_key][joint_key]) row = [camera_key, frame_key, joint_key, '%.3f' % e[camera_key][frame_key][joint_key]['rmse'], '%.3f' % e[camera_key][frame_key][joint_key]['x_error'], '%.3f' % e[camera_key][frame_key][joint_key]['y_error']] table.add_row(row) x_avg = x_avg + e[camera_key][frame_key][joint_key]['x_error'] y_avg = y_avg + e[camera_key][frame_key][joint_key]['y_error'] rmse_avg = rmse_avg + e[camera_key][frame_key][joint_key]['rmse'] total_joints = total_joints + 1 x_avg = x_avg / total_joints y_avg = y_avg / total_joints rmse_avg = rmse_avg / total_joints table.add_row(['--------', '--------', '-----------', '--------', '--------', '--------']) avg_row = [camera_key, frame_key, Fore.BLUE + Style.BRIGHT + 'Average' + Fore.BLACK + Style.NORMAL, '%.3f' % rmse_avg, '%.3f' % x_avg, '%.3f' % y_avg] table.add_row(avg_row) table.add_row(['--------', '--------', '-----------', '--------', '--------', '--------']) print(table) table = PrettyTable(table_header) x_avg = 0 y_avg = 0 rmse_avg = 0 for frame_key, frame in camera['frames'].items(): total_frames = 0 x_avg = 0 y_avg = 0 rmse_avg = 0 for joint_key, joint in joint_correspondence.items(): joint_key_detected = joint['body25'] joint_detected = frame['joints'][joint_key_detected] if not joint_detected['valid']: # skip invalid joints continue x_avg = x_avg + e[camera_key][frame_key][joint_key]['x_error'] y_avg = y_avg + e[camera_key][frame_key][joint_key]['y_error'] rmse_avg = rmse_avg + e[camera_key][frame_key][joint_key]['rmse'] total_frames = total_frames + 1 x_avg = x_avg / total_frames y_avg = y_avg / total_frames rmse_avg = rmse_avg / total_frames avg_row = [camera_key, Fore.BLUE + Style.BRIGHT + frame_key + Fore.BLACK + Style.NORMAL, 'Average', '%.3f' % rmse_avg, '%.3f' % x_avg, '%.3f' % y_avg] table.add_row(avg_row) print(table) table = PrettyTable(table_header) for joint_key, joint in joint_correspondence.items(): total_frames = 0 x_avg = 0 y_avg = 0 rmse_avg = 0 for frame_key, frame in camera['frames'].items(): joint_key_detected = joint['body25'] joint_detected = frame['joints'][joint_key_detected] if not joint_detected['valid']: # skip invalid joints continue x_avg = x_avg + e[camera_key][frame_key][joint_key]['x_error'] y_avg = y_avg + e[camera_key][frame_key][joint_key]['y_error'] rmse_avg = rmse_avg + e[camera_key][frame_key][joint_key]['rmse'] total_frames = total_frames + 1 x_avg = x_avg / total_frames y_avg = y_avg / total_frames rmse_avg = rmse_avg / total_frames avg_row = [camera_key, Fore.BLUE + Style.BRIGHT + 'Average' + Fore.BLACK + Style.NORMAL, joint_key, '%.3f' % rmse_avg, '%.3f' % x_avg, '%.3f' % y_avg] table.add_row(avg_row) print(table) if __name__ == "__main__": main()