zoom_diff_range = list(np.arange(0.2, 0.257, 0.001)) + list(
np.arange(0.257, 0.2, -0.005))
# iterate through all files
for idx, pc_filename in enumerate(pc_filenames):
# read point-cloud
velo_pc = read_velo_bin(pc_filename)
# read corresponding image
fname, file_ext = os.path.splitext(pc_filename)
fname = fname.split('/')[-1]
img_fname = os.path.join(args.img_dir, fname + '.png')
img_bgr = cv2.imread(img_fname)
img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
# perform prediction
pred_tuple, dt = trainer.predict(velo_pc, img_rgb)
if args.dense_depth:
label_dict, dense_depth = pred_tuple
# tracking
label_tracked_dict = []
if TRACKING == True:
# get detection list for mot_tracker
mot_det = []
mot_other_info = []
for label_ in label_dict:
det = [
label_['h'], label_['w'], label_['l'], label_['x'],
label_['y'], label_['z'], label_['yaw']
]
other_info = (label_['class'], label_['conf'])
train_depth_only=args.train_depth_only, train_obj_only=args.train_obj_only)
# show 100 samples
cv2.namedWindow('VR3Dense', cv2.WINDOW_NORMAL)
cv2.resizeWindow('VR3Dense', 800, 1440)
for i in range(100):
sample = trainer.dataset[i]
## get true labels visualization
pc_bbox_img_true = draw_point_cloud_w_bbox(sample['cloud'], sample['label_dict'], \
xlim=trainer.xlim, ylim=trainer.ylim, zlim=trainer.zlim)
pc_bbox_img_true_bgr = cv2.cvtColor(pc_bbox_img_true,
cv2.COLOR_RGB2BGR)
## get predicted labels visualization
# perform prediction
pred_tuple, dt = trainer.predict(sample['cloud'], sample['left_image'])
if args.dense_depth:
label_dict, dense_depth = pred_tuple
# get visualization
pc_bbox_img_pred = draw_point_cloud_w_bbox(sample['cloud'], label_dict, \
xlim=trainer.xlim, ylim=trainer.ylim, zlim=trainer.zlim)
pc_bbox_img_pred_bgr = cv2.cvtColor(pc_bbox_img_pred,
cv2.COLOR_RGB2BGR)
# visualization image
img_viz = cv2.vconcat([pc_bbox_img_true_bgr, pc_bbox_img_pred_bgr])
cv2.line(
img_viz, (0, pc_bbox_img_true_bgr.shape[0]),
(pc_bbox_img_true_bgr.shape[1] - 1, pc_bbox_img_true_bgr.shape[0]),
color=(255, 255, 255),
