''' Prepare KITTI data for 3D object detection.
Author: Charles R. Qi
Date: September 2017
'''
from __future__ import print_function
import os
import sys
import numpy as np
import cv2
from PIL import Image
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
ROOT_DIR = os.path.dirname(BASE_DIR)
sys.path.append(BASE_DIR)
sys.path.append(os.path.join(ROOT_DIR, 'mayavi'))
import kitti_util as utils
import cPickle as pickle
from kitti_object import *
import argparse
def in_hull(p, hull):
from scipy.spatial import Delaunay
if not isinstance(hull,Delaunay):
hull = Delaunay(hull)
return hull.find_simplex(p)>=0
def extract_pc_in_box3d(pc, box3d):
''' pc: (N,3), box3d: (8,3) '''
box3d_roi_inds = in_hull(pc[:,0:3], box3d)
return pc[box3d_roi_inds,:], box3d_roi_inds
def extract_pc_in_box2d(pc, box2d):
''' pc: (N,2), box2d: (xmin,ymin,xmax,ymax) '''
box2d_corners = np.zeros((4,2))
box2d_corners[0,:] = [box2d[0],box2d[1]]
box2d_corners[1,:] = [box2d[2],box2d[1]]
box2d_corners[2,:] = [box2d[2],box2d[3]]
box2d_corners[3,:] = [box2d[0],box2d[3]]
box2d_roi_inds = in_hull(pc[:,0:2], box2d_corners)
return pc[box2d_roi_inds,:], box2d_roi_inds
def demo():
import mayavi.mlab as mlab
from viz_util import draw_lidar, draw_lidar_simple, draw_gt_boxes3d
dataset = kitti_object(os.path.join(ROOT_DIR, 'dataset/KITTI/object'))
data_idx = 0
# Load data from dataset
objects = dataset.get_label_objects(data_idx)
objects[0].print_object()
img = dataset.get_image(data_idx)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img_height, img_width, img_channel = img.shape
print(('Image shape: ', img.shape))
pc_velo = dataset.get_lidar(data_idx)[:,0:3]
calib = dataset.get_calibration(data_idx)
## Draw lidar in rect camera coord
#print(' -------- LiDAR points in rect camera coordination --------')
#pc_rect = calib.project_velo_to_rect(pc_velo)
#fig = draw_lidar_simple(pc_rect)
#raw_input()
# Draw 2d and 3d boxes on image
print(' -------- 2D/3D bounding boxes in images --------')
show_image_with_boxes(img, objects, calib)
raw_input()
# Show all LiDAR points. Draw 3d box in LiDAR point cloud
print(' -------- LiDAR points and 3D boxes in velodyne coordinate --------')
#show_lidar_with_boxes(pc_velo, objects, calib)
#raw_input()
show_lidar_with_boxes(pc_velo, objects, calib, True, img_width, img_height)
raw_input()
# Visualize LiDAR points on images
print(' -------- LiDAR points projected to image plane --------')
show_lidar_on_image(pc_velo, img, calib, img_width, img_height)
raw_input()
# Show LiDAR points that are in the 3d box
print(' -------- LiDAR points in a 3D bounding box --------')
box3d_pts_2d, box3d_pts_3d = utils.compute_box_3d(objects[0], calib.P)
box3d_pts_3d_velo = calib.project_rect_to_velo(box3d_pts_3d)
box3droi_pc_velo, _ = extract_pc_in_box3d(pc_velo, box3d_pts_3d_velo)
print(('Number of points in 3d box: ', box3droi_pc_velo.shape[0]))
fig = mlab.figure(figure=None, bgcolor=(0,0,0),
fgcolor=None, engine=None, size=(1000, 500))
draw_lidar(box3droi_pc_velo, fig=fig)
draw_gt_boxes3d([box3d_pts_3d_velo], fig=fig)
mlab.show(1)
raw_input()
# UVDepth Image and its backprojection to point clouds
print(' -------- LiDAR points in a frustum from a 2D box --------')
imgfov_pc_velo, pts_2d, fov_inds = get_lidar_in_image_fov(pc_velo,
calib, 0, 0, img_width, img_height, True)
imgfov_pts_2d = pts_2d[fov_inds,:]
imgfov_pc_rect = calib.project_velo_to_rect(imgfov_pc_velo)
cameraUVDepth = np.zeros_like(imgfov_pc_rect)
cameraUVDepth[:,0:2] = imgfov_pts_2d
cameraUVDepth[:,2] = imgfov_pc_rect[:,2]
# Show that the points are exactly the same
backprojected_pc_velo = calib.project_image_to_velo(cameraUVDepth)
print(imgfov_pc_velo[0:20])
print(backprojected_pc_velo[0:20])
fig = mlab.figure(figure=None, bgcolor=(0,0,0),
fgcolor=None, engine=None, size=(1000, 500))
draw_lidar(backprojected_pc_velo, fig=fig)
raw_input()
# Only display those points that fall into 2d box
print(' -------- LiDAR points in a frustum from a 2D box --------')
xmin,ymin,xmax,ymax = \
objects[0].xmin, objects[0].ymin, objects[0].xmax, objects[0].ymax
boxfov_pc_velo = \
get_lidar_in_image_fov(pc_velo, calib, xmin, ymin, xmax, ymax)
print(('2d box FOV point num: ', boxfov_pc_velo.shape[0]))
fig = mlab.figure(figure=None, bgcolor=(0,0,0),
fgcolor=None, engine=None, size=(1000, 500))
draw_lidar(boxfov_pc_velo, fig=fig)
mlab.show(1)
raw_input()
def random_shift_box2d(box2d, shift_ratio=0.1):
''' Randomly shift box center, randomly scale width and height
'''
r = shift_ratio
xmin,ymin,xmax,ymax = box2d
h = ymax-ymin
w = xmax-xmin
cx = (xmin+xmax)/2.0
cy = (ymin+ymax)/2.0
cx2 = cx + w*r*(np.random.random()*2-1)
cy2 = cy + h*r*(np.random.random()*2-1)
h2 = h*(1+np.random.random()*2*r-r) # 0.9 to 1.1
w2 = w*(1+np.random.random()*2*r-r) # 0.9 to 1.1
return np.array([cx2-w2/2.0, cy2-h2/2.0, cx2+w2/2.0, cy2+h2/2.0])
def extract_frustum_data(idx_filename, split, output_filename, viz=False,
perturb_box2d=False, augmentX=1, type_whitelist=['Car']):
''' Extract point clouds and corresponding annotations in frustums
defined generated from 2D bounding boxes
Lidar points and 3d boxes are in *rect camera* coord system
(as that in 3d box label files)
Input:
idx_filename: string, each line of the file is a sample ID
split: string, either trianing or testing
output_filename: string, the name for output .pickle file
viz: bool, whether to visualize extracted data
perturb_box2d: bool, whether to perturb the box2d
(used for data augmentation in train set)
augmentX: scalar, how many augmentations to have for each 2D box.
type_whitelist: a list of strings, object types we are interested in.
Output:
None (will write a .pickle file to the disk)
'''
dataset = kitti_object(os.path.join(ROOT_DIR,'dataset/KITTI/object'), split)
data_idx_list = [int(line.rstrip()) for line in open(idx_filename)]
id_list = [] # int number
box2d_list = [] # [xmin,ymin,xmax,ymax]
box3d_list = [] # (8,3) array in rect camera coord
input_list = [] # channel number = 4, xyz,intensity in rect camera coord
label_list = [] # 1 for roi object, 0 for clutter
type_list = [] # string e.g. Car
heading_list = [] # ry (along y-axis in rect camera coord) radius of
# (cont.) clockwise angle from positive x axis in velo coord.
box3d_size_list = [] # array of l,w,h
frustum_angle_list = [] # angle of 2d box center from pos x-axis
pos_cnt = 0
all_cnt = 0
for data_idx in data_idx_list:
print('------------- ', data_idx)
calib = dataset.get_calibration(data_idx) # 3 by 4 matrix
objects = dataset.get_label_objects(data_idx)
pc_velo = dataset.get_lidar(data_idx)
pc_rect = np.zeros_like(pc_velo)
pc_rect[:,0:3] = calib.project_velo_to_rect(pc_velo[:,0:3])
pc_rect[:,3] = pc_velo[:,3]
img = dataset.get_image(data_idx)
img_height, img_width, img_channel = img.shape
_, pc_image_coord, img_fov_inds = get_lidar_in_image_fov(pc_velo[:,0:3],
calib, 0, 0, img_width, img_height, True)
for obj_idx in range(len(objects)):
if objects[obj_idx].type not in type_whitelist :continue
# 2D BOX: Get pts rect backprojected
box2d = objects[obj_idx].box2d
for _ in range(augmentX):
# Augment data by box2d perturbation
if perturb_box2d:
xmin,ymin,xmax,ymax = random_shift_box2d(box2d)
print(box2d)
print(xmin,ymin,xmax,ymax)
else:
xmin,ymin,xmax,ymax = box2d
box_fov_inds = (pc_image_coord[:,0]<xmax) & \
(pc_image_coord[:,0]>=xmin) & \
(pc_image_coord[:,1]<ymax) & \
(pc_image_coord[:,1]>=ymin)
box_fov_inds = box_fov_inds & img_fov_inds
pc_in_box_fov = pc_rect[box_fov_inds,:]
# Get frustum angle (according to center pixel in 2D BOX)
box2d_center = np.array([(xmin+xmax)/2.0, (ymin+ymax)/2.0])
uvdepth = np.zeros((1,3))
uvdepth[0,0:2] = box2d_center
uvdepth[0,2] = 20 # some random depth
box2d_center_rect = calib.project_image_to_rect(uvdepth)
frustum_angle = -1 * np.arctan2(box2d_center_rect[0,2],
box2d_center_rect[0,0])
# 3D BOX: Get pts velo in 3d box
obj = objects[obj_idx]
box3d_pts_2d, box3d_pts_3d = utils.compute_box_3d(obj, calib.P)
_,inds = extract_pc_in_box3d(pc_in_box_fov, box3d_pts_3d)
label = np.zeros((pc_in_box_fov.shape[0]))
label[inds] = 1
# Get 3D BOX heading
heading_angle = obj.ry
# Get 3D BOX size
box3d_size = np.array([obj.l, obj.w, obj.h])
# Reject too far away object or object without points
if ymax-ymin<25 or np.sum(label)==0:
continue
id_list.append(data_idx)
box2d_list.append(np.array([xmin,ymin,xmax,ymax]))
box3d_list.append(box3d_pts_3d)
input_list.append(pc_in_box_fov)
label_list.append(label)
type_list.append(objects[obj_idx].type)
heading_list.append(heading_angle)
box3d_size_list.append(box3d_size)
frustum_angle_list.append(frustum_angle)
# collect statistics
pos_cnt += np.sum(label)
all_cnt += pc_in_box_fov.shape[0]
print('Average pos ratio: %f' % (pos_cnt/float(all_cnt)))
print('Average npoints: %f' % (float(all_cnt)/len(id_list)))
with open(output_filename,'wb') as fp:
pickle.dump(id_list, fp)
pickle.dump(box2d_list,fp)
pickle.dump(box3d_list,fp)
pickle.dump(input_list, fp)
pickle.dump(label_list, fp)
pickle.dump(type_list, fp)
pickle.dump(heading_list, fp)
pickle.dump(box3d_size_list, fp)
pickle.dump(frustum_angle_list, fp)
if viz:
import mayavi.mlab as mlab
for i in range(10):
p1 = input_list[i]
seg = label_list[i]
fig = mlab.figure(figure=None, bgcolor=(0.4,0.4,0.4),
fgcolor=None, engine=None, size=(500, 500))
mlab.points3d(p1[:,0], p1[:,1], p1[:,2], seg, mode='point',
colormap='gnuplot', scale_factor=1, figure=fig)
fig = mlab.figure(figure=None, bgcolor=(0.4,0.4,0.4),
fgcolor=None, engine=None, size=(500, 500))
mlab.points3d(p1[:,2], -p1[:,0], -p1[:,1], seg, mode='point',
colormap='gnuplot', scale_factor=1, figure=fig)
raw_input()
def get_box3d_dim_statistics(idx_filename):
''' Collect and dump 3D bounding box statistics '''
dataset = kitti_object(os.path.join(ROOT_DIR,'dataset/KITTI/object'))
dimension_list = []
type_list = []
ry_list = []
data_idx_list = [int(line.rstrip()) for line in open(idx_filename)]
for data_idx in data_idx_list:
print('------------- ', data_idx)
calib = dataset.get_calibration(data_idx) # 3 by 4 matrix
objects = dataset.get_label_objects(data_idx)
for obj_idx in range(len(objects)):
obj = objects[obj_idx]
if obj.type=='DontCare':continue
dimension_list.append(np.array([obj.l,obj.w,obj.h]))
type_list.append(obj.type)
ry_list.append(obj.ry)
with open('box3d_dimensions.pickle','wb') as fp:
pickle.dump(type_list, fp)
pickle.dump(dimension_list, fp)
pickle.dump(ry_list, fp)
def read_det_file(det_filename):
''' Parse lines in 2D detection output files '''
det_id2str = {1:'Pedestrian', 2:'Car', 3:'Cyclist'}
id_list = []
type_list = []
prob_list = []
box2d_list = []
for line in open(det_filename, 'r'):
t = line.rstrip().split(" ")
id_list.append(int(os.path.basename(t[0]).rstrip('.png')))
type_list.append(det_id2str[int(t[1])])
prob_list.append(float(t[2]))
box2d_list.append(np.array([float(t[i]) for i in range(3,7)]))
return id_list, type_list, box2d_list, prob_list
def extract_frustum_data_rgb_detection(det_filename, split, output_filename,
viz=False,
type_whitelist=['Car'],
img_height_threshold=25,
lidar_point_threshold=5):
''' Extract point clouds in frustums extruded from 2D detection boxes.
Update: Lidar points and 3d boxes are in *rect camera* coord system
(as that in 3d box label files)
Input:
det_filename: string, each line is
img_path typeid confidence xmin ymin xmax ymax
split: string, either trianing or testing
output_filename: string, the name for output .pickle file
type_whitelist: a list of strings, object types we are interested in.
img_height_threshold: int, neglect image with height lower than that.
lidar_point_threshold: int, neglect frustum with too few points.
Output:
None (will write a .pickle file to the disk)
'''
dataset = kitti_object(os.path.join(ROOT_DIR, 'dataset/KITTI/object'), split)
det_id_list, det_type_list, det_box2d_list, det_prob_list = \
read_det_file(det_filename)
cache_id = -1
cache = None
id_list = []
type_list = []
box2d_list = []
prob_list = []
input_list = [] # channel number = 4, xyz,intensity in rect camera coord
frustum_angle_list = [] # angle of 2d box center from pos x-axis
for det_idx in range(len(det_id_list)):
data_idx = det_id_list[det_idx]
print('det idx: %d/%d, data idx: %d' % \
(det_idx, len(det_id_list), data_idx))
if cache_id != data_idx:
calib = dataset.get_calibration(data_idx) # 3 by 4 matrix
pc_velo = dataset.get_lidar(data_idx)
pc_rect = np.zeros_like(pc_velo)
pc_rect[:,0:3] = calib.project_velo_to_rect(pc_velo[:,0:3])
pc_rect[:,3] = pc_velo[:,3]
img = dataset.get_image(data_idx)
img_height, img_width, img_channel = img.shape
_, pc_image_coord, img_fov_inds = get_lidar_in_image_fov(\
pc_velo[:,0:3], calib, 0, 0, img_width, img_height, True)
cache = [calib,pc_rect,pc_image_coord,img_fov_inds]
cache_id = data_idx
else:
calib,pc_rect,pc_image_coord,img_fov_inds = cache
if det_type_list[det_idx] not in type_whitelist: continue
# 2D BOX: Get pts rect backprojected
xmin,ymin,xmax,ymax = det_box2d_list[det_idx]
box_fov_inds = (pc_image_coord[:,0]<xmax) & \
(pc_image_coord[:,0]>=xmin) & \
(pc_image_coord[:,1]<ymax) & \
(pc_image_coord[:,1]>=ymin)
box_fov_inds = box_fov_inds & img_fov_inds
pc_in_box_fov = pc_rect[box_fov_inds,:]
# Get frustum angle (according to center pixel in 2D BOX)
box2d_center = np.array([(xmin+xmax)/2.0, (ymin+ymax)/2.0])
uvdepth = np.zeros((1,3))
uvdepth[0,0:2] = box2d_center
uvdepth[0,2] = 20 # some random depth
box2d_center_rect = calib.project_image_to_rect(uvdepth)
frustum_angle = -1 * np.arctan2(box2d_center_rect[0,2],
box2d_center_rect[0,0])
# Pass objects that are too small
if ymax-ymin<img_height_threshold or \
len(pc_in_box_fov)<lidar_point_threshold:
continue
id_list.append(data_idx)
type_list.append(det_type_list[det_idx])
box2d_list.append(det_box2d_list[det_idx])
prob_list.append(det_prob_list[det_idx])
input_list.append(pc_in_box_fov)
frustum_angle_list.append(frustum_angle)
with open(output_filename,'wb') as fp:
pickle.dump(id_list, fp)
pickle.dump(box2d_list,fp)
pickle.dump(input_list, fp)
pickle.dump(type_list, fp)
pickle.dump(frustum_angle_list, fp)
pickle.dump(prob_list, fp)
if viz:
import mayavi.mlab as mlab
for i in range(10):
p1 = input_list[i]
fig = mlab.figure(figure=None, bgcolor=(0.4,0.4,0.4),
fgcolor=None, engine=None, size=(500, 500))
mlab.points3d(p1[:,0], p1[:,1], p1[:,2], p1[:,1], mode='point',
colormap='gnuplot', scale_factor=1, figure=fig)
fig = mlab.figure(figure=None, bgcolor=(0.4,0.4,0.4),
fgcolor=None, engine=None, size=(500, 500))
mlab.points3d(p1[:,2], -p1[:,0], -p1[:,1], seg, mode='point',
colormap='gnuplot', scale_factor=1, figure=fig)
raw_input()
def write_2d_rgb_detection(det_filename, split, result_dir):
''' Write 2D detection results for KITTI evaluation.
Convert from Wei's format to KITTI format.
Input:
det_filename: string, each line is
img_path typeid confidence xmin ymin xmax ymax
split: string, either trianing or testing
result_dir: string, folder path for results dumping
Output:
None (will write <xxx>.txt files to disk)
Usage:
write_2d_rgb_detection("val_det.txt", "training", "results")
'''
dataset = kitti_object(os.path.join(ROOT_DIR, 'dataset/KITTI/object'), split)
det_id_list, det_type_list, det_box2d_list, det_prob_list = \
read_det_file(det_filename)
# map from idx to list of strings, each string is a line without \n
results = {}
for i in range(len(det_id_list)):
idx = det_id_list[i]
typename = det_type_list[i]
box2d = det_box2d_list[i]
prob = det_prob_list[i]
output_str = typename + " -1 -1 -10 "
output_str += "%f %f %f %f " % (box2d[0],box2d[1],box2d[2],box2d[3])
output_str += "-1 -1 -1 -1000 -1000 -1000 -10 %f" % (prob)
if idx not in results: results[idx] = []
results[idx].append(output_str)
if not os.path.exists(result_dir): os.mkdir(result_dir)
output_dir = os.path.join(result_dir, 'data')
if not os.path.exists(output_dir): os.mkdir(output_dir)
for idx in results:
pred_filename = os.path.join(output_dir, '%06d.txt'%(idx))
fout = open(pred_filename, 'w')
for line in results[idx]:
fout.write(line+'\n')
fout.close()
if __name__=='__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--demo', action='store_true', help='Run demo.')
parser.add_argument('--gen_train', action='store_true', help='Generate train split frustum data with perturbed GT 2D boxes')
parser.add_argument('--gen_val', action='store_true', help='Generate val split frustum data with GT 2D boxes')
parser.add_argument('--gen_val_rgb_detection', action='store_true', help='Generate val split frustum data with RGB detection 2D boxes')
parser.add_argument('--car_only', action='store_true', help='Only generate cars; otherwise cars, peds and cycs')
args = parser.parse_args()
if args.demo:
demo()
exit()
if args.car_only:
type_whitelist = ['Car']
output_prefix = 'frustum_caronly_'
else:
type_whitelist = ['Car', 'Pedestrian', 'Cyclist']
output_prefix = 'frustum_carpedcyc_'
if args.gen_train:
extract_frustum_data(\
os.path.join(BASE_DIR, 'image_sets/train.txt'),
'training',
os.path.join(BASE_DIR, output_prefix+'train.pickle'),
viz=False, perturb_box2d=True, augmentX=5,
type_whitelist=type_whitelist)
if args.gen_val:
extract_frustum_data(\
os.path.join(BASE_DIR, 'image_sets/val.txt'),
'training',
os.path.join(BASE_DIR, output_prefix+'val.pickle'),
viz=False, perturb_box2d=False, augmentX=1,
type_whitelist=type_whitelist)
if args.gen_val_rgb_detection:
extract_frustum_data_rgb_detection(\
os.path.join(BASE_DIR, 'rgb_detections/rgb_detection_val.txt'),
'training',
os.path.join(BASE_DIR, output_prefix+'val_rgb_detection.pickle'),
viz=False,
type_whitelist=type_whitelist)
