def generate_rpn_training_labels(pts_rect, gt_boxes3d):
cls_label = np.zeros((pts_rect.shape[0]), dtype=np.int32)
reg_label = np.zeros((pts_rect.shape[0], 7),
dtype=np.float32) # dx, dy, dz, ry, h, w, l
gt_corners = kitti_utils.boxes3d_to_corners3d(gt_boxes3d, rotate=True)
extend_gt_boxes3d = kitti_utils.enlarge_box3d(gt_boxes3d, extra_width=0.2)
extend_gt_corners = kitti_utils.boxes3d_to_corners3d(extend_gt_boxes3d,
rotate=True)
for k in range(gt_boxes3d.shape[0]):
box_corners = gt_corners[k]
fg_pt_flag = in_hull(pts_rect, box_corners)
fg_pts_rect = pts_rect[fg_pt_flag]
cls_label[fg_pt_flag] = 1
# enlarge the bbox3d, ignore nearby points
extend_box_corners = extend_gt_corners[k]
fg_enlarge_flag = in_hull(pts_rect, extend_box_corners)
ignore_flag = np.logical_xor(fg_pt_flag, fg_enlarge_flag)
cls_label[ignore_flag] = -1
# pixel offset of object center
center3d = gt_boxes3d[k][0:3].copy() # (x, y, z)
center3d[1] -= gt_boxes3d[k][3] / 2
reg_label[
fg_pt_flag, 0:
3] = center3d - fg_pts_rect # Now y is the true center of 3d box 20180928
# size and angle encoding
reg_label[fg_pt_flag, 3] = gt_boxes3d[k][3] # h
reg_label[fg_pt_flag, 4] = gt_boxes3d[k][4] # w
reg_label[fg_pt_flag, 5] = gt_boxes3d[k][5] # l
reg_label[fg_pt_flag, 6] = gt_boxes3d[k][6] # ry
return cls_label, reg_label
def aug_roi_by_noise(self, roi_info):
"""
add noise to original roi to get aug_box3d
:param roi_info:
:return:
"""
roi_box3d, gt_box3d = roi_info['roi_box3d'], roi_info['gt_box3d']
original_iou = roi_info['iou3d']
temp_iou = cnt = 0
pos_thresh = min(cfg.RCNN.REG_FG_THRESH, cfg.RCNN.CLS_FG_THRESH)
gt_corners = kitti_utils.boxes3d_to_corners3d(gt_box3d.reshape(-1, 7))
aug_box3d = roi_box3d
while temp_iou < pos_thresh and cnt < 10:
if roi_info['type'] == 'gt':
aug_box3d = self.random_aug_box3d(roi_box3d) # GT, must random
else:
if np.random.rand() < 0.2:
aug_box3d = roi_box3d # p=0.2 to keep the original roi box
else:
aug_box3d = self.random_aug_box3d(roi_box3d)
aug_corners = kitti_utils.boxes3d_to_corners3d(aug_box3d.reshape(-1, 7))
iou3d = kitti_utils.get_iou3d(aug_corners, gt_corners)
temp_iou = iou3d[0][0]
cnt += 1
if original_iou < pos_thresh: # original bg, break
break
return aug_box3d
def aug_roi_by_noise_batch(self, roi_boxes3d, gt_boxes3d, aug_times=10):
"""
:param roi_boxes3d: (N, 7)
:param gt_boxes3d: (N, 7)
:return:
"""
iou_of_rois = np.zeros(roi_boxes3d.shape[0], dtype=np.float32)
for k in range(roi_boxes3d.__len__()):
temp_iou = cnt = 0
roi_box3d = roi_boxes3d[k]
gt_box3d = gt_boxes3d[k]
pos_thresh = min(cfg.RCNN.REG_FG_THRESH, cfg.RCNN.CLS_FG_THRESH)
gt_corners = kitti_utils.boxes3d_to_corners3d(gt_box3d.reshape(1, 7))
aug_box3d = roi_box3d
while temp_iou < pos_thresh and cnt < aug_times:
if np.random.rand() < 0.2:
aug_box3d = roi_box3d # p=0.2 to keep the original roi box
else:
aug_box3d = self.random_aug_box3d(roi_box3d)
aug_corners = kitti_utils.boxes3d_to_corners3d(aug_box3d.reshape(1, 7))
iou3d = kitti_utils.get_iou3d(aug_corners, gt_corners)
temp_iou = iou3d[0][0]
cnt += 1
roi_boxes3d[k] = aug_box3d
iou_of_rois[k] = temp_iou
return roi_boxes3d, iou_of_rois
def save_kitti_format(calib, bbox3d, obj_list, img_shape, save_fp):
corners3d = kitti_utils.boxes3d_to_corners3d(bbox3d)
img_boxes, _ = calib.corners3d_to_img_boxes(corners3d)
img_boxes[:, 0] = np.clip(img_boxes[:, 0], 0, img_shape[1] - 1)
img_boxes[:, 1] = np.clip(img_boxes[:, 1], 0, img_shape[0] - 1)
img_boxes[:, 2] = np.clip(img_boxes[:, 2], 0, img_shape[1] - 1)
img_boxes[:, 3] = np.clip(img_boxes[:, 3], 0, img_shape[0] - 1)
# Discard boxes that are larger than 80% of the image width OR height
img_boxes_w = img_boxes[:, 2] - img_boxes[:, 0]
img_boxes_h = img_boxes[:, 3] - img_boxes[:, 1]
box_valid_mask = np.logical_and(img_boxes_w < img_shape[1] * 0.8, img_boxes_h < img_shape[0] * 0.8)
for k in range(bbox3d.shape[0]):
if box_valid_mask[k] == 0:
continue
x, z, ry = bbox3d[k, 0], bbox3d[k, 2], bbox3d[k, 6]
beta = np.arctan2(z, x)
alpha = -np.sign(beta) * np.pi / 2 + beta + ry
print('%s %.2f %d %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f' %
(args.class_name, obj_list[k].trucation, int(obj_list[k].occlusion), alpha, img_boxes[k, 0], img_boxes[k, 1],
img_boxes[k, 2], img_boxes[k, 3],
bbox3d[k, 3], bbox3d[k, 4], bbox3d[k, 5], bbox3d[k, 0], bbox3d[k, 1], bbox3d[k, 2],
bbox3d[k, 6]), file=save_fp)
def save_kitti_format(sample_id, calib, bbox3d, kitti_output_dir, scores, img_shape):
corners3d = kitti_utils.boxes3d_to_corners3d(bbox3d)
img_boxes, _ = calib.corners3d_to_img_boxes(corners3d)
img_boxes[:, 0] = np.clip(img_boxes[:, 0], 0, img_shape[1] - 1)
img_boxes[:, 1] = np.clip(img_boxes[:, 1], 0, img_shape[0] - 1)
img_boxes[:, 2] = np.clip(img_boxes[:, 2], 0, img_shape[1] - 1)
img_boxes[:, 3] = np.clip(img_boxes[:, 3], 0, img_shape[0] - 1)
img_boxes_w = img_boxes[:, 2] - img_boxes[:, 0]
img_boxes_h = img_boxes[:, 3] - img_boxes[:, 1]
box_valid_mask = np.logical_and(img_boxes_w < img_shape[1] * 0.8, img_boxes_h < img_shape[0] * 0.8)
kitti_output_file = os.path.join(kitti_output_dir, '%06d.txt' % sample_id)
with open(kitti_output_file, 'w') as f:
for k in range(bbox3d.shape[0]):
if box_valid_mask[k] == 0:
continue
x, z, ry = bbox3d[k, 0], bbox3d[k, 2], bbox3d[k, 6]
beta = np.arctan2(z, x)
alpha = -np.sign(beta) * np.pi / 2 + beta + ry
##### cfg.CLASSES 대신에 bbox3d[k, 7]의 string version를 넣어야한다
print('%s -1 -1 %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f' %
(cfg.CLASSES, alpha, img_boxes[k, 0], img_boxes[k, 1], img_boxes[k, 2], img_boxes[k, 3],
bbox3d[k, 3], bbox3d[k, 4], bbox3d[k, 5], bbox3d[k, 0], bbox3d[k, 1], bbox3d[k, 2],
bbox3d[k, 6], scores[k]), file=f)
def dist_to_boxes(points, boxes):
"""
Calculates combined distance for each point to all boxes
:param points: (N, 3)
:param boxes: (N, 7) [x, y, z, h, w, l, ry]
:return: distances_array: (M) torch.Tensor of [(N), (N), ...] distances
"""
distances_array = torch.Tensor([])
box_corners = kitti_utils.boxes3d_to_corners3d(boxes)
for box in box_corners:
minX = min(box[:, 0])
minY = min(box[:, 1])
minZ = min(box[:, 2])
maxX = max(box[:, 0])
maxY = max(box[:, 1])
maxZ = max(box[:, 2])
centroid = np.array([(maxX + minX) / 2, (maxY + minY) / 2,
(maxZ + minZ) / 2])
dists_to_curr_box = dist_to_box_centroid(
torch.from_numpy(points),
torch.from_numpy(centroid)).reshape(1, len(points))
distances_array = torch.cat(
(distances_array.float(), dists_to_curr_box.float()), 0)
return distances_array
def save_argo_format(sample_id, bbox3d, argo_output_dir, scores,
lidar_idx_table):
data = []
corners3d = kitti_utils.boxes3d_to_corners3d(bbox3d)
argo_output_file = os.path.join(
argo_output_dir, lidar_idx_table['%06d' % sample_id] + '.json')
with open(argo_output_file, 'w') as f:
for k in range(bbox3d.shape[0]):
temp = {}
temp['center'] = {
'x': float(bbox3d[k, 2]),
'y': -float(bbox3d[k, 0]),
'z': -float(bbox3d[k, 1])
}
temp['length'] = float(bbox3d[k, 5])
temp['width'] = float(bbox3d[k, 4])
temp['height'] = float(bbox3d[k, 3])
q = pyquaternion.Quaternion(axis=(0.0, 0.0, 1.0),
radians=(np.pi / 2. - bbox3d[k, 6]))
temp['rotation'] = {'w': q.w, 'x': q.x, 'y': q.y, 'z': q.z}
temp['score'] = float(scores[k])
temp['label_class'] = cfg.CLASSES
data.append(temp)
json.dump(data, f)
def save_kitti_format(sample_id, bbox3d, kitti_output_dir, scores,
lidar_idx_table):
corners3d = kitti_utils.boxes3d_to_corners3d(bbox3d)
print("--------------------")
kitti_output_file = os.path.join(
kitti_output_dir, lidar_idx_table['%06d' % sample_id] + '.txt')
with open(kitti_output_file, 'w') as f:
print(f)
for k in range(bbox3d.shape[0]):
x, z, ry = bbox3d[k, 0], bbox3d[k, 2], bbox3d[k, 6]
beta = np.arctan2(z, x)
alpha = -np.sign(beta) * np.pi / 2 + beta + ry
print('%s %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f' %
(cfg.CLASSES, bbox3d[k, 3], bbox3d[k, 4], bbox3d[k, 5],
bbox3d[k, 2], -bbox3d[k, 0], -bbox3d[k, 1],
(np.pi / 2. - bbox3d[k, 6]), scores[k]),
file=f)
def visualize_lidar_plane(self, bbox3d, frame_id):
marker_array = MarkerArray()
marker = Marker()
marker.header.frame_id = frame_id
marker.type = marker.LINE_LIST
marker.action = marker.ADD
marker.header.stamp = rospy.Time.now()
# marker scale (scale y and z not used due to being linelist)
marker.scale.x = 0.08
# marker color
marker.color.a = 1.0
marker.color.r = 1.0
marker.color.g = 1.0
marker.color.b = 0.0
marker.pose.position.x = 0.0
marker.pose.position.y = 0.0
marker.pose.position.z = 0.0
marker.pose.orientation.x = 0.0
marker.pose.orientation.y = 0.0
marker.pose.orientation.z = 0.0
marker.pose.orientation.w = 1.0
marker.points = []
corner_for_box_list = [0, 1, 0, 3, 2, 3, 2, 1, 4, 5, 4, 7, 6, 7, 6, 5, 3, 7, 0, 4, 1, 5, 2, 6]
corners3d = kitti_utils.boxes3d_to_corners3d(bbox3d) # (N,8,3)
for box_nr in range(corners3d.shape[0]):
box3d_pts_3d_velo = corners3d[box_nr] # (8,3)
for corner in corner_for_box_list:
p = np.array(box3d_pts_3d_velo[corner, 0:4])
transformed_p = transform_point(p, np.linalg.inv(self.Tr_velo_kitti_cam))
p = Point()
p.x = transformed_p[0]
p.y = transformed_p[1]
p.z = transformed_p[2]
marker.points.append(p)
marker_array.markers.append(marker)
id = 0
for m in marker_array.markers:
m.id = id
id += 1
self.mk_pub.publish(marker_array)
marker_array.markers = []
pass
for label in labels:
x_list.append(label.pos[0])
y_list.append(label.pos[1])
z_list.append(label.pos[2])
l_list.append(label.l)
w_list.append(label.w)
h_list.append(label.h)
ry_list.append(label.ry)
points = align_img_and_pc(img_dir, pc_dir, calib_dir)
# print("numpoints: ", len(points))
num_pts_per_scene.append(len(points))
# Get the foreground and background label
bboxes3d = kitti_utils.objs_to_boxes3d(labels)
# print("Number of bboxes: ",len(bboxes3d))
bboxes3d_rotated_corners = kitti_utils.boxes3d_to_corners3d(bboxes3d)
box3d_roi_inds_overall = None
sub_box3d_roi_inds_overall = None
valid_labels = []
for i, bbox3d_corners in enumerate(bboxes3d_rotated_corners):
# print("bboxes3d_rotated_corners: ", bboxes3d_rotated_corners[i])
box3d_roi_inds = kitti_utils.in_hull(points[:, :3], bbox3d_corners)
# box3d_roi_inds = kitti_utils.in_hull(bbox3d_corners[:,:3], bbox3d_corners)
# print("xmin: ", np.min(points[:,0]), " xmax: ", np.max(points[:,0]))
# print("ymin: ", np.min(points[:,1]), " ymax: ", np.max(points[:,1]))
# print("zmin: ", np.min(points[:,2]), " zmax: ", np.max(points[:,2]))
# pc_filter = kitti_utils.extract_pc_in_box3d(points[:,:3], bbox3d_corners)
# sub_pc_filter = kitti_utils.extract_pc_in_box3d(points[:,:3], bbox3d_corners)
# print("pc_filter.shape: ", pc_filter[0].shape)
# print("interested indices shape: ", box3d_roi_inds.shape)
# print("interested indices: ", box3d_roi_inds)
def get_rcnn_training_sample_batch(self, index):
sample_id = int(self.sample_id_list[index])
rpn_xyz, rpn_features, rpn_intensity, seg_mask = \
self.get_rpn_features(self.rcnn_training_feature_dir, sample_id)
# load rois and gt_boxes3d for this sample
roi_file = os.path.join(self.rcnn_training_roi_dir, '%06d.txt' % sample_id)
roi_obj_list = kitti_utils.get_objects_from_label(roi_file)
roi_boxes3d = kitti_utils.objs_to_boxes3d(roi_obj_list)
# roi_scores = kitti_utils.objs_to_scores(roi_obj_list)
gt_obj_list = self.filtrate_objects(self.get_label(sample_id))
gt_boxes3d = kitti_utils.objs_to_boxes3d(gt_obj_list)
# calculate original iou
iou3d = kitti_utils.get_iou3d(kitti_utils.boxes3d_to_corners3d(roi_boxes3d),
kitti_utils.boxes3d_to_corners3d(gt_boxes3d))
max_overlaps, gt_assignment = iou3d.max(axis=1), iou3d.argmax(axis=1)
max_iou_of_gt, roi_assignment = iou3d.max(axis=0), iou3d.argmax(axis=0)
roi_assignment = roi_assignment[max_iou_of_gt > 0].reshape(-1)
# sample fg, easy_bg, hard_bg
fg_rois_per_lidar = int(np.round(cfg.RCNN.FG_RATIO * cfg.RCNN.ROI_PER_lidar))
fg_thresh = min(cfg.RCNN.REG_FG_THRESH, cfg.RCNN.CLS_FG_THRESH)
fg_inds = np.nonzero(max_overlaps >= fg_thresh)[0]
fg_inds = np.concatenate((fg_inds, roi_assignment), axis=0) # consider the roi which has max_overlaps with gt as fg
easy_bg_inds = np.nonzero((max_overlaps < cfg.RCNN.CLS_BG_THRESH_LO))[0]
hard_bg_inds = np.nonzero((max_overlaps < cfg.RCNN.CLS_BG_THRESH) &
(max_overlaps >= cfg.RCNN.CLS_BG_THRESH_LO))[0]
fg_num_rois = fg_inds.size
bg_num_rois = hard_bg_inds.size + easy_bg_inds.size
if fg_num_rois > 0 and bg_num_rois > 0:
# sampling fg
fg_rois_per_this_lidar = min(fg_rois_per_lidar, fg_num_rois)
rand_num = np.random.permutation(fg_num_rois)
fg_inds = fg_inds[rand_num[:fg_rois_per_this_lidar]]
# sampling bg
bg_rois_per_this_lidar = cfg.RCNN.ROI_PER_lidar - fg_rois_per_this_lidar
bg_inds = self.sample_bg_inds(hard_bg_inds, easy_bg_inds, bg_rois_per_this_lidar)
elif fg_num_rois > 0 and bg_num_rois == 0:
# sampling fg
rand_num = np.floor(np.random.rand(cfg.RCNN.ROI_PER_lidar ) * fg_num_rois)
rand_num = torch.from_numpy(rand_num).type_as(gt_boxes3d).long()
fg_inds = fg_inds[rand_num]
fg_rois_per_this_lidar = cfg.RCNN.ROI_PER_lidar
bg_rois_per_this_lidar = 0
elif bg_num_rois > 0 and fg_num_rois == 0:
# sampling bg
bg_rois_per_this_lidar = cfg.RCNN.ROI_PER_lidar
bg_inds = self.sample_bg_inds(hard_bg_inds, easy_bg_inds, bg_rois_per_this_lidar)
fg_rois_per_this_lidar = 0
else:
import pdb
pdb.set_trace()
raise NotImplementedError
# augment the rois by noise
roi_list, roi_iou_list, roi_gt_list = [], [], []
if fg_rois_per_this_lidar > 0:
fg_rois_src = roi_boxes3d[fg_inds].copy()
gt_of_fg_rois = gt_boxes3d[gt_assignment[fg_inds]]
fg_rois, fg_iou3d = self.aug_roi_by_noise_batch(fg_rois_src, gt_of_fg_rois, aug_times=10)
roi_list.append(fg_rois)
roi_iou_list.append(fg_iou3d)
roi_gt_list.append(gt_of_fg_rois)
if bg_rois_per_this_lidar > 0:
bg_rois_src = roi_boxes3d[bg_inds].copy()
gt_of_bg_rois = gt_boxes3d[gt_assignment[bg_inds]]
bg_rois, bg_iou3d = self.aug_roi_by_noise_batch(bg_rois_src, gt_of_bg_rois, aug_times=1)
roi_list.append(bg_rois)
roi_iou_list.append(bg_iou3d)
roi_gt_list.append(gt_of_bg_rois)
rois = np.concatenate(roi_list, axis=0)
iou_of_rois = np.concatenate(roi_iou_list, axis=0)
gt_of_rois = np.concatenate(roi_gt_list, axis=0)
# collect extra features for point cloud pooling
if cfg.RCNN.USE_INTENSITY:
pts_extra_input_list = [rpn_intensity.reshape(-1, 1), seg_mask.reshape(-1, 1)]
else:
pts_extra_input_list = [seg_mask.reshape(-1, 1)]
if cfg.RCNN.USE_DEPTH:
pts_depth = (np.linalg.norm(rpn_xyz, ord=2, axis=1) / 70.0) - 0.5
pts_extra_input_list.append(pts_depth.reshape(-1, 1))
pts_extra_input = np.concatenate(pts_extra_input_list, axis=1)
pts_input, pts_features, pts_empty_flag = roipool3d_utils.roipool3d_cpu(rois, rpn_xyz, rpn_features,
pts_extra_input,
cfg.RCNN.POOL_EXTRA_WIDTH,
sampled_pt_num=cfg.RCNN.NUM_POINTS,
canonical_transform=False)
valid_mask = (pts_empty_flag == 0).astype(np.int32)
# regression valid mask
reg_valid_mask = (iou_of_rois > cfg.RCNN.REG_FG_THRESH).astype(np.int32) & valid_mask
# classification label
cls_label = (iou_of_rois > cfg.RCNN.CLS_FG_THRESH).astype(np.int32)
invalid_mask = (iou_of_rois > cfg.RCNN.CLS_BG_THRESH) & (iou_of_rois < cfg.RCNN.CLS_FG_THRESH)
cls_label[invalid_mask] = -1
cls_label[valid_mask == 0] = -1
# canonical transform and sampling
pts_input_ct, gt_boxes3d_ct = self.canonical_transform_batch(pts_input, rois, gt_of_rois)
sample_info = {'sample_id': sample_id,
'pts_input': pts_input_ct,
'pts_features': pts_features,
'cls_label': cls_label,
'reg_valid_mask': reg_valid_mask,
'gt_boxes3d_ct': gt_boxes3d_ct,
'roi_boxes3d': rois,
'roi_size': rois[:, 3:6],
'gt_boxes3d': gt_of_rois}
return sample_info
def get_rcnn_sample_info(self, roi_info):
sample_id, gt_box3d = roi_info['sample_id'], roi_info['gt_box3d']
rpn_xyz, rpn_features, rpn_intensity, seg_mask = self.rpn_feature_list[sample_id]
# augmentation original roi by adding noise
roi_box3d = self.aug_roi_by_noise(roi_info)
# point cloud pooling based on roi_box3d
pooled_boxes3d = kitti_utils.enlarge_box3d(roi_box3d.reshape(1, 7), cfg.RCNN.POOL_EXTRA_WIDTH)
boxes_pts_mask_list = roipool3d_utils.pts_in_boxes3d_cpu(torch.from_numpy(rpn_xyz),
torch.from_numpy(pooled_boxes3d))
pt_mask_flag = (boxes_pts_mask_list[0].numpy() == 1)
cur_pts = rpn_xyz[pt_mask_flag].astype(np.float32)
# data augmentation
aug_pts = cur_pts.copy()
aug_gt_box3d = gt_box3d.copy().astype(np.float32)
aug_roi_box3d = roi_box3d.copy()
if cfg.AUG_DATA and self.mode == 'TRAIN':
# calculate alpha by ry
temp_boxes3d = np.concatenate([aug_roi_box3d.reshape(1, 7), aug_gt_box3d.reshape(1, 7)], axis=0)
temp_x, temp_z, temp_ry = temp_boxes3d[:, 0], temp_boxes3d[:, 2], temp_boxes3d[:, 6]
temp_beta = np.arctan2(temp_z, temp_x).astype(np.float64)
temp_alpha = -np.sign(temp_beta) * np.pi / 2 + temp_beta + temp_ry
# data augmentation
aug_pts, aug_boxes3d, aug_method = self.data_augmentation(aug_pts, temp_boxes3d, temp_alpha, mustaug=True, stage=2)
aug_roi_box3d, aug_gt_box3d = aug_boxes3d[0], aug_boxes3d[1]
aug_gt_box3d = aug_gt_box3d.astype(gt_box3d.dtype)
# Pool input points
valid_mask = 1 # whether the input is valid
if aug_pts.shape[0] == 0:
pts_features = np.zeros((1, 128), dtype=np.float32)
input_channel = 3 + int(cfg.RCNN.USE_INTENSITY) + int(cfg.RCNN.USE_MASK) + int(cfg.RCNN.USE_DEPTH)
pts_input = np.zeros((1, input_channel), dtype=np.float32)
valid_mask = 0
else:
pts_features = rpn_features[pt_mask_flag].astype(np.float32)
pts_intensity = rpn_intensity[pt_mask_flag].astype(np.float32)
pts_input_list = [aug_pts, pts_intensity.reshape(-1, 1)]
if cfg.RCNN.USE_INTENSITY:
pts_input_list = [aug_pts, pts_intensity.reshape(-1, 1)]
else:
pts_input_list = [aug_pts]
if cfg.RCNN.USE_MASK:
if cfg.RCNN.MASK_TYPE == 'seg':
pts_mask = seg_mask[pt_mask_flag].astype(np.float32)
elif cfg.RCNN.MASK_TYPE == 'roi':
pts_mask = roipool3d_utils.pts_in_boxes3d_cpu(torch.from_numpy(aug_pts),
torch.from_numpy(aug_roi_box3d.reshape(1, 7)))
pts_mask = (pts_mask[0].numpy() == 1).astype(np.float32)
else:
raise NotImplementedError
pts_input_list.append(pts_mask.reshape(-1, 1))
if cfg.RCNN.USE_DEPTH:
pts_depth = np.linalg.norm(aug_pts, axis=1, ord=2)
pts_depth_norm = (pts_depth / 70.0) - 0.5
pts_input_list.append(pts_depth_norm.reshape(-1, 1))
pts_input = np.concatenate(pts_input_list, axis=1) # (N, C)
aug_gt_corners = kitti_utils.boxes3d_to_corners3d(aug_gt_box3d.reshape(-1, 7))
aug_roi_corners = kitti_utils.boxes3d_to_corners3d(aug_roi_box3d.reshape(-1, 7))
iou3d = kitti_utils.get_iou3d(aug_roi_corners, aug_gt_corners)
cur_iou = iou3d[0][0]
# regression valid mask
reg_valid_mask = 1 if cur_iou >= cfg.RCNN.REG_FG_THRESH and valid_mask == 1 else 0
# classification label
cls_label = 1 if cur_iou > cfg.RCNN.CLS_FG_THRESH else 0
if cfg.RCNN.CLS_BG_THRESH < cur_iou < cfg.RCNN.CLS_FG_THRESH or valid_mask == 0:
cls_label = -1
# canonical transform and sampling
pts_input_ct, gt_box3d_ct = self.canonical_transform(pts_input, aug_roi_box3d, aug_gt_box3d)
pts_input_ct, pts_features = self.rcnn_input_sample(pts_input_ct, pts_features)
sample_info = {'sample_id': sample_id,
'pts_input': pts_input_ct,
'pts_features': pts_features,
'cls_label': cls_label,
'reg_valid_mask': reg_valid_mask,
'gt_boxes3d_ct': gt_box3d_ct,
'roi_boxes3d': aug_roi_box3d,
'roi_size': aug_roi_box3d[3:6],
'gt_boxes3d': aug_gt_box3d}
return sample_info
def apply_gt_aug_to_one_scene(self, sample_id, pts_rect, pts_intensity, all_gt_boxes3d):
"""
:param pts_rect: (N, 3)
:param all_gt_boxex3d: (M2, 7)
:return:
"""
assert self.gt_database is not None
# extra_gt_num = np.random.randint(10, 15)
# try_times = 50
if cfg.GT_AUG_RAND_NUM:
extra_gt_num = np.random.randint(10, cfg.GT_EXTRA_NUM)
else:
extra_gt_num = cfg.GT_EXTRA_NUM
try_times = 100
cnt = 0
cur_gt_boxes3d = all_gt_boxes3d.copy()
cur_gt_boxes3d[:, 4] += 0.5 # TODO: consider different objects
cur_gt_boxes3d[:, 5] += 0.5 # enlarge new added box to avoid too nearby boxes
cur_gt_corners = kitti_utils.boxes3d_to_corners3d(cur_gt_boxes3d)
extra_gt_obj_list = []
extra_gt_boxes3d_list = []
new_pts_list, new_pts_intensity_list = [], []
src_pts_flag = np.ones(pts_rect.shape[0], dtype=np.int32)
road_plane = self.get_road_plane(sample_id)
a, b, c, d = road_plane
while try_times > 0:
if cnt > extra_gt_num:
break
try_times -= 1
if cfg.GT_AUG_HARD_RATIO > 0:
p = np.random.rand()
if p > cfg.GT_AUG_HARD_RATIO:
# use easy sample
rand_idx = np.random.randint(0, len(self.gt_database[0]))
new_gt_dict = self.gt_database[0][rand_idx]
else:
# use hard sample
rand_idx = np.random.randint(0, len(self.gt_database[1]))
new_gt_dict = self.gt_database[1][rand_idx]
else:
rand_idx = np.random.randint(0, self.gt_database.__len__())
new_gt_dict = self.gt_database[rand_idx]
new_gt_box3d = new_gt_dict['gt_box3d'].copy()
new_gt_points = new_gt_dict['points'].copy()
new_gt_intensity = new_gt_dict['intensity'].copy()
new_gt_obj = new_gt_dict['obj']
center = new_gt_box3d[0:3]
if cfg.PC_REDUCE_BY_RANGE and (self.check_pc_range(center) is False):
continue
if new_gt_points.__len__() < 5: # too few points
continue
# put it on the road plane
cur_height = (-d - a * center[0] - c * center[2]) / b
move_height = new_gt_box3d[1] - cur_height
new_gt_box3d[1] -= move_height
new_gt_points[:, 1] -= move_height
new_gt_obj.pos[1] -= move_height
new_enlarged_box3d = new_gt_box3d.copy()
new_enlarged_box3d[4] += 0.5
new_enlarged_box3d[5] += 0.5 # enlarge new added box to avoid too nearby boxes
cnt += 1
new_corners = kitti_utils.boxes3d_to_corners3d(new_enlarged_box3d.reshape(1, 7))
iou3d = kitti_utils.get_iou3d(new_corners, cur_gt_corners)
valid_flag = iou3d.max() < 1e-8
if not valid_flag:
continue
enlarged_box3d = new_gt_box3d.copy()
enlarged_box3d[3] += 2 # remove the points above and below the object
boxes_pts_mask_list = roipool3d_utils.pts_in_boxes3d_cpu(
torch.from_numpy(pts_rect), torch.from_numpy(enlarged_box3d.reshape(1, 7)))
pt_mask_flag = (boxes_pts_mask_list[0].numpy() == 1)
src_pts_flag[pt_mask_flag] = 0 # remove the original points which are inside the new box
new_pts_list.append(new_gt_points)
new_pts_intensity_list.append(new_gt_intensity)
cur_gt_boxes3d = np.concatenate((cur_gt_boxes3d, new_enlarged_box3d.reshape(1, 7)), axis=0)
cur_gt_corners = np.concatenate((cur_gt_corners, new_corners), axis=0)
extra_gt_boxes3d_list.append(new_gt_box3d.reshape(1, 7))
extra_gt_obj_list.append(new_gt_obj)
if new_pts_list.__len__() == 0:
return False, pts_rect, pts_intensity, None, None
extra_gt_boxes3d = np.concatenate(extra_gt_boxes3d_list, axis=0)
# remove original points and add new points
pts_rect = pts_rect[src_pts_flag == 1]
pts_intensity = pts_intensity[src_pts_flag == 1]
new_pts_rect = np.concatenate(new_pts_list, axis=0)
new_pts_intensity = np.concatenate(new_pts_intensity_list, axis=0)
pts_rect = np.concatenate((pts_rect, new_pts_rect), axis=0)
pts_intensity = np.concatenate((pts_intensity, new_pts_intensity), axis=0)
return True, pts_rect, pts_intensity, extra_gt_boxes3d, extra_gt_obj_list
box_id = -1
fg_flag = False
gt_box = np.zeros(7)
gt_mask = np.zeros((cur_prob_mask.shape))
if foreground_flag[i] == True:
fg_flag = True
if foreground_flag_G[i] == True:
box_id = proposal_gt_index[i].detach().cpu().numpy()
gt_box = gt_boxes3d_cam[box_id].detach().cpu().numpy().reshape(7)
gt_box[0] = gt_box[0] - cur_pts_center[0]
gt_box[2] = gt_box[2] - cur_pts_center[2]
gt_box[3] = gt_box[3] * 1.2
gt_box[4] = gt_box[4] * 1.2
gt_box[5] = gt_box[5] * 1.2
gt_corners = kitti_utils.boxes3d_to_corners3d(gt_box.reshape(
-1, 7),
rotate=True)
gt_mask = kitti_utils.in_hull(cur_box_point,
gt_corners.reshape(-1, 3)).reshape(
-1, 1)
gt_box = gt_boxes3d_cam[box_id].detach().cpu().numpy().reshape(7)
gt_box[0] = gt_box[0] - cur_pts_center[0]
gt_box[2] = gt_box[2] - cur_pts_center[2]
# fig, ax = plt.subplots(figsize=(5, 5))
# plt.title('%d / %d'%(box_id, gt_boxes3d_cam.shape[0]))
# ax.axis([-4, 4, -4, 4])
# plt.scatter(cur_box_point[:, 0], cur_box_point[:, 2], s=15, c=cur_prob_mask[:,0], edgecolor='none',
# cmap=plt.get_cmap('rainbow'), alpha=1, marker='.', vmin=0, vmax=1)
# plt.scatter(np.zeros(1), np.zeros(1), s=200, c='black',
def get_proposal_from_file(self, index):
sample_id = int(self.lidar_idx_list[index])
proposal_file = os.path.join(self.rcnn_eval_roi_dir, '%06d.txt' % sample_id)
roi_obj_list = kitti_utils.get_objects_from_label(proposal_file)
rpn_xyz, rpn_features, rpn_intensity, seg_mask = self.get_rpn_features(self.rcnn_eval_feature_dir, sample_id)
pts_rect, pts_rpn_features, pts_intensity = rpn_xyz, rpn_features, rpn_intensity
roi_box3d_list, roi_scores = [], []
for obj in roi_obj_list:
box3d = np.array([obj.pos[0], obj.pos[1], obj.pos[2], obj.h, obj.w, obj.l, obj.ry], dtype=np.float32)
roi_box3d_list.append(box3d.reshape(1, 7))
roi_scores.append(obj.score)
roi_boxes3d = np.concatenate(roi_box3d_list, axis=0) # (N, 7)
roi_scores = np.array(roi_scores, dtype=np.float32) # (N)
if cfg.RCNN.ROI_SAMPLE_JIT:
sample_dict = {'sample_id': sample_id,
'rpn_xyz': rpn_xyz,
'rpn_features': rpn_features,
'seg_mask': seg_mask,
'roi_boxes3d': roi_boxes3d,
'roi_scores': roi_scores,
'pts_depth': np.linalg.norm(rpn_xyz, ord=2, axis=1)}
if self.mode != 'TEST':
gt_obj_list = self.filtrate_objects(self.get_label(sample_id))
gt_boxes3d = kitti_utils.objs_to_boxes3d(gt_obj_list)
roi_corners = kitti_utils.boxes3d_to_corners3d(roi_boxes3d)
gt_corners = kitti_utils.boxes3d_to_corners3d(gt_boxes3d)
iou3d = kitti_utils.get_iou3d(roi_corners, gt_corners)
if gt_boxes3d.shape[0] > 0:
gt_iou = iou3d.max(axis=1)
else:
gt_iou = np.zeros(roi_boxes3d.shape[0]).astype(np.float32)
sample_dict['gt_boxes3d'] = gt_boxes3d
sample_dict['gt_iou'] = gt_iou
return sample_dict
if cfg.RCNN.USE_INTENSITY:
pts_extra_input_list = [pts_intensity.reshape(-1, 1), seg_mask.reshape(-1, 1)]
else:
pts_extra_input_list = [seg_mask.reshape(-1, 1)]
if cfg.RCNN.USE_DEPTH:
cur_depth = np.linalg.norm(pts_rect, axis=1, ord=2)
cur_depth_norm = (cur_depth / 70.0) - 0.5
pts_extra_input_list.append(cur_depth_norm.reshape(-1, 1))
pts_extra_input = np.concatenate(pts_extra_input_list, axis=1)
pts_input, pts_features = roipool3d_utils.roipool3d_cpu(roi_boxes3d, pts_rect, pts_rpn_features,
pts_extra_input, cfg.RCNN.POOL_EXTRA_WIDTH,
sampled_pt_num=cfg.RCNN.NUM_POINTS)
sample_dict = {'sample_id': sample_id,
'pts_input': pts_input,
'pts_features': pts_features,
'roi_boxes3d': roi_boxes3d,
'roi_scores': roi_scores,
'roi_size': roi_boxes3d[:, 3:6]}
if self.mode == 'TEST':
return sample_dict
gt_obj_list = self.filtrate_objects(self.get_label(sample_id))
gt_boxes3d = np.zeros((gt_obj_list.__len__(), 7), dtype=np.float32)
for k, obj in enumerate(gt_obj_list):
gt_boxes3d[k, 0:3], gt_boxes3d[k, 3], gt_boxes3d[k, 4], gt_boxes3d[k, 5], gt_boxes3d[k, 6] \
= obj.pos, obj.h, obj.w, obj.l, obj.ry
if gt_boxes3d.__len__() == 0:
gt_iou = np.zeros((roi_boxes3d.shape[0]), dtype=np.float32)
else:
roi_corners = kitti_utils.boxes3d_to_corners3d(roi_boxes3d)
gt_corners = kitti_utils.boxes3d_to_corners3d(gt_boxes3d)
iou3d = kitti_utils.get_iou3d(roi_corners, gt_corners)
gt_iou = iou3d.max(axis=1)
sample_dict['gt_boxes3d'] = gt_boxes3d
sample_dict['gt_iou'] = gt_iou
return sample_dict
def save_kitti_format(sample_id, calib, bbox3d, kitti_output_dir, scores,
img_shape):
corners3d = kitti_utils.boxes3d_to_corners3d(bbox3d)
img_boxes, _ = calib.corners3d_to_img_boxes(corners3d)
# frame number cumsum for kitti_tracking dataset, from 0000 to 0020
img_count_cumsum = [
154, 597, 830, 974, 1288, 1585, 1855, 2655, 3045, 3848, 4142, 4515,
4593, 4933, 5039, 5415, 5624, 5769, 6108, 7167, 8004
]
def convert_idx(idx, image_count_cumsum):
# print(image_count_cumsum)
for i in range(len(image_count_cumsum)):
if idx >= image_count_cumsum[i]:
continue
else:
if i == 0:
frame_id_in_video = idx
else:
frame_id_in_video = idx - image_count_cumsum[i - 1]
frame_id = frame_id_in_video
video_id = i
break
return video_id, frame_id
video_id, frame_id = convert_idx(sample_id, img_count_cumsum)
img_boxes[:, 0] = np.clip(img_boxes[:, 0], 0, img_shape[1] - 1)
img_boxes[:, 1] = np.clip(img_boxes[:, 1], 0, img_shape[0] - 1)
img_boxes[:, 2] = np.clip(img_boxes[:, 2], 0, img_shape[1] - 1)
img_boxes[:, 3] = np.clip(img_boxes[:, 3], 0, img_shape[0] - 1)
img_boxes_w = img_boxes[:, 2] - img_boxes[:, 0]
img_boxes_h = img_boxes[:, 3] - img_boxes[:, 1]
box_valid_mask = np.logical_and(img_boxes_w < img_shape[1] * 0.8,
img_boxes_h < img_shape[0] * 0.8)
# kitti_output_file = os.path.join(kitti_output_dir, '%06d.txt' % sample_id)
kitti_output_file = os.path.join(kitti_output_dir, '%04d.txt' % video_id)
# with open(kitti_output_file, 'w') as f:
with open(kitti_output_file, 'a') as f:
for k in range(bbox3d.shape[0]):
if box_valid_mask[k] == 0:
continue
x, z, ry = bbox3d[k, 0], bbox3d[k, 2], bbox3d[k, 6]
beta = np.arctan2(z, x)
alpha = -np.sign(beta) * np.pi / 2 + beta + ry
# print('%s -1 -1 %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f' %
# (cfg.CLASSES, alpha, img_boxes[k, 0], img_boxes[k, 1], img_boxes[k, 2], img_boxes[k, 3],
# bbox3d[k, 3], bbox3d[k, 4], bbox3d[k, 5], bbox3d[k, 0], bbox3d[k, 1], bbox3d[k, 2],
# bbox3d[k, 6], scores[k]), file=f)
print(
'%d,2,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f,%.4f'
% (frame_id, img_boxes[k, 0], img_boxes[k, 1], img_boxes[k, 2],
img_boxes[k, 3], scores[k], bbox3d[k, 3], bbox3d[k, 4],
bbox3d[k, 5], bbox3d[k, 0], bbox3d[k, 1], bbox3d[k, 2],
bbox3d[k, 6], alpha),
file=f)
def eval_one_epoch_joint(model, dataloader, epoch_id, result_dir, logger):
np.random.seed(666)
MEAN_SIZE = torch.from_numpy(cfg.CLS_MEAN_SIZE[0]).cuda()
mode = 'TEST' if args.test else 'EVAL'
final_output_dir = os.path.join(result_dir, 'final_result', 'data')
if os.path.exists(final_output_dir): shutil.rmtree(final_output_dir)
os.makedirs(final_output_dir, exist_ok=True)
logger.info('---- EPOCH %s JOINT EVALUATION ----' % epoch_id)
logger.info('==> Output file: %s' % result_dir)
model.eval()
thresh_list = [0.1, 0.3, 0.5, 0.7, 0.9]
total_recalled_bbox_list, total_gt_bbox = [0] * 5, 0
total_roi_recalled_bbox_list = [0] * 5
dataset = dataloader.dataset
cnt = final_total = total_cls_acc = total_cls_acc_refined = total_rpn_iou = 0
obj_num = 0
progress_bar = tqdm.tqdm(total=len(dataloader), leave=True, desc='eval')
iou_list = []
iou_p_score_list = []
rcnn_p_score_list = []
prop_count = 0
for data in dataloader:
# Loading sample
sample_id_list, pts_input = data['sample_id'], data['pts_input']
sample_id = sample_id_list[0]
cnt += len(sample_id_list)
#if cnt < 118: continue
#load label
if not args.test:
gt_boxes3d = data['gt_boxes3d']
obj_num += gt_boxes3d.shape[1]
# print(obj_num)
if gt_boxes3d.shape[1] == 0: # (B, M, 7)
pass
else:
gt_boxes3d = gt_boxes3d
# rpn model inference
inputs = torch.from_numpy(pts_input).cuda(non_blocking=True).float()
#inputs = inputs[:,torch.argsort(-inputs[0,:,2])]
input_data = {'pts_input': inputs}
ret_dict = model.rpn_forward(input_data)
rpn_cls, rpn_reg = ret_dict['rpn_cls'], ret_dict['rpn_reg']
rpn_backbone_xyz, rpn_backbone_features = ret_dict[
'backbone_xyz'], ret_dict['backbone_features']
# stage score parsing
rpn_scores_raw = rpn_cls[:, :, 0]
rpn_scores_norm = torch.sigmoid(rpn_cls[:, :, 0])
rcnn_input_scores = rpn_scores_norm.view(-1).clone()
inputs = inputs.view(-1, inputs.shape[-1])
rpn_backbone_features = rpn_backbone_features.view(
-1, rpn_backbone_features.shape[-2])
rpn_backbone_xyz = rpn_backbone_xyz.view(-1,
rpn_backbone_xyz.shape[-1])
# if VISUAL:
# order = torch.argsort(-rpn_scores_norm).view(-1)
# inputs = inputs.view(-1,inputs.shape[-1])[order]
# rpn_scores_norm = rpn_scores_norm.view(-1)[order]
# rpn_backbone_features = rpn_backbone_features.view(-1,rpn_backbone_features.shape[-1])[order]
#
# norm_feature = F.normalize(rpn_backbone_features)
# similarity = norm_feature.mm(norm_feature.t())
#
# inputs_plt = inputs.detach().cpu().numpy()
# scores_plt = rpn_scores_norm.detach().cpu().numpy()
# similarity_plt = similarity.detach().cpu().numpy()
#
#
# fig = plt.figure(figsize=(10, 10))
# plt.axes(facecolor='silver')
# plt.axis([-30,30,0,70])
# plt.title('point_regressed_center %06d'%sample_id)
# plt.scatter(inputs_plt[:, 0], inputs_plt[:, 2], s=15, c=scores_plt[:], edgecolor='none',
# cmap=plt.get_cmap('rainbow'), alpha=1, marker='.', vmin=0, vmax=1)
# if args.test==False:
# gt_boxes3d = gt_boxes3d.reshape(-1,7)
# plt.scatter(gt_boxes3d[:, 0], gt_boxes3d[:, 2], s=200, c='blue',
# alpha=0.5, marker='+', vmin=-1, vmax=1)
# plt.show()
#
# for i in range(similarity_plt.shape[0]):
# fig = plt.figure(figsize=(10, 10))
# plt.axes(facecolor='silver')
# plt.axis([-30, 30, 0, 70])
# sm_plt = similarity_plt[i]
# plt.scatter(inputs_plt[i, 0].reshape(-1), inputs_plt[i, 2].reshape(-1), s=400, c='blue',
# alpha=0.5, marker='+', vmin=-1, vmax=1)
# plt.scatter(inputs_plt[:, 0], inputs_plt[:, 2], s=15, c=(sm_plt[:]+scores_plt[:])/2, edgecolor='none',
# cmap=plt.get_cmap('rainbow'), alpha=1, marker='.', vmin=0, vmax=1)
# plt.show()
# thresh select and jump out
# rpn_mask = rpn_scores_norm.view(-1) > cfg.RPN.SCORE_THRESH
# if rpn_mask.float().sum() == 0: continue
# rpn_scores_raw = rpn_scores_raw.view(-1)[rpn_mask]
# rpn_scores_norm = rpn_scores_norm.view(-1)[rpn_mask]
# rpn_reg = rpn_reg.view(-1, rpn_reg.shape[-1])[rpn_mask]
# rpn_backbone_xyz = rpn_backbone_xyz.view(-1, rpn_backbone_xyz.shape[-1])[rpn_mask]
# generate rois
rpn_rois = decode_center_target(
rpn_backbone_xyz,
rpn_reg.view(-1, rpn_reg.shape[-1]),
loc_scope=cfg.RPN.LOC_SCOPE,
loc_bin_size=cfg.RPN.LOC_BIN_SIZE,
).view(-1, 3)
rpn_reg_dist = (rpn_rois - rpn_backbone_xyz).clone()
#similarity = torch.cosine_similarity(rpn_backbone_xyz[:, [0, 2]], rpn_reg_dist[:, [0, 2]], dim=1)
# # thresh select and jump out
rpn_mask = (rpn_scores_norm.view(-1) > cfg.RPN.SCORE_THRESH) & (
rpn_reg_dist[:, [0, 2]].pow(2).sum(-1).sqrt() > 0.2) #\
#& (similarity > -0.7)
if rpn_mask.float().sum() == 0: continue
rpn_scores_raw = rpn_scores_raw.view(-1)[rpn_mask]
rpn_scores_norm = rpn_scores_norm.view(-1)[rpn_mask]
rpn_rois = rpn_rois[rpn_mask]
rpn_backbone_xyz = rpn_backbone_xyz.view(
-1, rpn_backbone_xyz.shape[-1])[rpn_mask]
# radius NMS
# sort by center score
sort_points = torch.argsort(-rpn_scores_raw)
rpn_rois = rpn_rois[sort_points]
rpn_scores_norm = rpn_scores_norm[sort_points]
rpn_scores_raw = rpn_scores_raw[sort_points]
if rpn_rois.shape[0] > 1:
keep_id = [0]
prop_prop_distance = distance_2(rpn_rois[:, [0, 2]],
rpn_rois[:, [0, 2]])
for i in range(1, rpn_rois.shape[0]):
#if torch.min(prop_prop_distance[:i, i], dim=-1)[0] > 0.3:
if torch.min(prop_prop_distance[keep_id, i], dim=-1)[0] > 0.3:
keep_id.append(i)
rpn_center = rpn_rois[keep_id][:, [0, 2]]
rpn_scores_norm = rpn_scores_norm[keep_id]
rpn_scores_raw = rpn_scores_raw[keep_id]
else:
rpn_center = rpn_rois[:, [0, 2]]
rpn_scores_norm = rpn_scores_norm
rpn_scores_raw = rpn_scores_raw
# #rcnn input select:
point_center_distance = distance_2(rpn_center, inputs[:, [0, 2]])
cur_proposal_points_index = (torch.min(point_center_distance,
dim=-1)[0] < 4.0)
point_center_distance = point_center_distance[
cur_proposal_points_index]
inputs = inputs[cur_proposal_points_index]
rcnn_input_scores = rcnn_input_scores.view(
-1)[cur_proposal_points_index]
if VISUAL:
inputs_plt = inputs.detach().cpu().numpy()
scores_plt = rcnn_input_scores.detach().cpu().numpy()
# point_center= rpn_center[rpn_scores_norm > 0.5]
# point_center_score = rpn_scores_norm[rpn_scores_norm > 0.5]
point_center = rpn_center
point_center_score = rpn_scores_norm
fig = plt.figure(figsize=(10, 10))
plt.axes(facecolor='silver')
plt.axis([-30, 30, 0, 70])
point_center_plt = point_center.cpu().numpy()
plt.title('point_regressed_center %06d' % sample_id)
plt.scatter(inputs_plt[:, 0],
inputs_plt[:, 2],
s=15,
c=scores_plt[:],
edgecolor='none',
cmap=plt.get_cmap('rainbow'),
alpha=1,
marker='.',
vmin=0,
vmax=1)
if point_center.shape[0] > 0:
plt.scatter(point_center_plt[:, 0],
point_center_plt[:, 1],
s=200,
c='white',
alpha=0.5,
marker='x',
vmin=-1,
vmax=1)
if args.test == False:
gt_boxes3d = gt_boxes3d.reshape(-1, 7)
plt.scatter(gt_boxes3d[:, 0],
gt_boxes3d[:, 2],
s=200,
c='blue',
alpha=0.5,
marker='+',
vmin=-1,
vmax=1)
plt.savefig('../visual/rpn.jpg')
# RCNN stage
box_list = []
raw_score_list = []
iou_score_list = []
inputs[:, 1] -= 1.65
point_center_distance = distance_2(rpn_center[:, :], inputs[:, [0, 2]])
#for c in range(min(rpn_center.shape[0],100)):
prop_count += rpn_center.shape[0]
print('num %d' % (prop_count / float(cnt)))
for c in range(rpn_center.shape[0]):
# rcnn input generate
cur_input = inputs.clone()
cur_input_score = rcnn_input_scores.clone()
# if COSINE_DISTANCE:
# cur_center_points_index = ((point_center_distance[:, c] < 4.0) & \
# (point_prop_cos_matrix[:, c] > COS_THRESH) | \
# (point_center_distance[:, c].view(-1) < 0.7)).view(-1)
# else:
cur_center_points_index = (point_center_distance[:, c] <
4.0).view(-1)
if cur_center_points_index.long().sum() == 0: continue
cur_center_points_xyz = cur_input[cur_center_points_index, :3]
cur_center_points_xyz[:, 0] -= rpn_center[c, 0]
cur_center_points_xyz[:, 2] -= rpn_center[c, 1]
cur_center_points_r = cur_input[cur_center_points_index,
3].view(-1, 1)
cur_center_points_mask = (cur_input_score[cur_center_points_index]
> 0.5).view(-1, 1).float()
# # easy sample sampling
# if pts_input.shape[0]>512:
# cur_input = torch.cat((cur_center_points_xyz, cur_center_points_r,
# (cur_input_score[cur_center_points_index] > 0.5).view(-1, 1).float()), dim=-1)
# pts_input = cur_input
# pts_input = pts_input[:min(pts_input.shape[0], 2000), :]
# pts_input = pts_input[:, :]
# sample_index = fps(pts_input[:, 0:3].contiguous(), ratio=min(512 / pts_input.shape[0], 0.99),
# random_start=False)
# perm = sample_index
# while sample_index.shape[0] < 512:
# sample_index = torch.cat(
# (sample_index, perm[:min(perm.shape[0], 512 - sample_index.shape[0])]), dim=0)
#
# cur_center_points_xyz = pts_input[sample_index, 0:3]
# cur_center_points_r = pts_input[sample_index, 3].reshape(-1, 1)
# cur_center_points_mask = pts_input[sample_index, 4].reshape(-1, 1)
cur_center_points_xyz = cur_center_points_xyz.unsqueeze(0).float()
cur_center_points_r = cur_center_points_r.unsqueeze(0).float()
cur_center_points_mask = cur_center_points_mask.unsqueeze(
0).float() - 0.5
input_data = {
'cur_box_point': cur_center_points_xyz,
'cur_box_reflect': cur_center_points_r,
'train_mask': cur_center_points_mask,
}
# # globaly random sampling
# pts_input = pts_input[:min(pts_input.shape[0], self.npoints), :]
# sample_index = np.arange(0, pts_input.shape[0], 1).astype(np.int)
# perm = np.copy(sample_index)
# while sample_index.shape[0] < self.npoints:
# sample_index = np.concatenate(
# (sample_index, perm[:min(perm.shape[0], self.npoints - sample_index.shape[0])]))
#
# cur_box_point = pts_input[sample_index, 0:3]
# cur_box_reflect = pts_input[sample_index, 3].reshape(-1, 1)
# cur_prob_mask = pts_input[sample_index, 4].reshape(-1, 1)
# gt_mask = pts_input[sample_index, 5].reshape(-1, 1)
# rcnn model inference
ret_dict = model.rcnn_forward(input_data)
rcnn_cls = ret_dict['rcnn_cls']
ioun_cls = ret_dict['ioun_cls']
rcnn_reg = ret_dict['rcnn_reg']
rcnn_iou = ret_dict['rcnn_iou']
rcnn_ref = ret_dict['rcnn_ref'].view(1, 1, -1)
rcnn_box3d = ret_dict['pred_boxes3d']
refined_box = ret_dict['refined_box']
rcnn_box3d = refined_box
rcnn_box3d[:, :, 6] = rcnn_box3d[:, :, 6] % (np.pi * 2)
if rcnn_box3d[:, :, 6] > np.pi: rcnn_box3d[:, :, 6] -= np.pi * 2
rcnn_box3d[:, :, 0] += rpn_center[c][0]
rcnn_box3d[:, :, 2] += rpn_center[c][1]
rcnn_box3d[:, :, 1] += 1.65
box_list.append(rcnn_box3d)
raw_score_list.append(rcnn_cls.view(1, 1))
#raw_score_list.append(ioun_cls.view(1,1))
iou_score_list.append(rcnn_iou.view(1, 1))
rcnn_box3d = torch.cat((box_list), dim=1)
raw_rcnn_score = torch.cat((raw_score_list),
dim=0).unsqueeze(0).float()
norm_ioun_score = torch.cat((iou_score_list),
dim=0).unsqueeze(0).float()
# scoring
pred_boxes3d = rcnn_box3d
norm_ioun_score = norm_ioun_score
raw_rcnn_score = raw_rcnn_score
norm_rcnn_score = torch.sigmoid(raw_rcnn_score)
# scores thresh
pred_h = pred_boxes3d[:, :, 3].view(-1)
pred_w = pred_boxes3d[:, :, 4].view(-1)
pred_l = pred_boxes3d[:, :, 5].view(-1)
inds = (norm_rcnn_score > cfg.RCNN.SCORE_THRESH) & (
norm_ioun_score > cfg.IOUN.SCORE_THRESH)
inds = inds.view(-1)
#size filiter
# inds = inds & \
# (pred_h > 1.2) & (pred_h < 2.2) & \
# (pred_w > 1.3) & (pred_w < 2.0) & \
# (pred_l > 2.2) & (pred_l < 5.0)
inds = inds & \
(pred_h > 1.1) & (pred_h < 2.3) & \
(pred_w > 1.2) & (pred_w < 2.1) & \
(pred_l > 2.1) & (pred_l < 5.1)
pred_boxes3d = pred_boxes3d[:, inds]
norm_rcnn_score = norm_rcnn_score[:, inds]
norm_ioun_score = norm_ioun_score[:, inds]
raw_rcnn_score = raw_rcnn_score[:, inds]
if pred_boxes3d.shape[1] == 0: continue
# evaluation
recalled_num = gt_num = 0
if not args.test:
gt_boxes3d = data['gt_boxes3d']
for k in range(1):
# calculate recall
cur_gt_boxes3d = gt_boxes3d[k]
tmp_idx = cur_gt_boxes3d.__len__() - 1
while tmp_idx >= 0 and cur_gt_boxes3d[tmp_idx].sum() == 0:
tmp_idx -= 1
if tmp_idx >= 0:
cur_gt_boxes3d = cur_gt_boxes3d[:tmp_idx + 1]
cur_gt_boxes3d = torch.from_numpy(cur_gt_boxes3d).cuda(
non_blocking=True).float()
_, iou3d = iou3d_utils.boxes_iou3d_gpu(
pred_boxes3d[k], cur_gt_boxes3d)
gt_max_iou, _ = iou3d.max(dim=0)
refined_iou, _ = iou3d.max(dim=1)
iou_list.append(refined_iou.view(-1, 1))
iou_p_score_list.append(norm_ioun_score.view(-1, 1))
rcnn_p_score_list.append(norm_rcnn_score.view(-1, 1))
for idx, thresh in enumerate(thresh_list):
total_recalled_bbox_list[idx] += (gt_max_iou >
thresh).sum().item()
recalled_num += (gt_max_iou > 0.7).sum().item()
gt_num += cur_gt_boxes3d.shape[0]
total_gt_bbox += cur_gt_boxes3d.shape[0]
if cnt == 1000:
iou_clloe = torch.cat(iou_list, dim=0).detach().cpu().numpy()
iou_score_clloe = torch.cat(iou_p_score_list,
dim=0).detach().cpu().numpy()
plt.axis([-.1, 1.1, -.1, 1.1])
plt.scatter(iou_clloe,
iou_score_clloe,
s=20,
c='blue',
edgecolor='none',
cmap=plt.get_cmap('YlOrRd'),
alpha=1,
marker='.')
plt.savefig(os.path.join(result_dir, 'distributercnn.png'))
disp_dict = {
'mode': mode,
'recall': '%d/%d' % (total_recalled_bbox_list[3], total_gt_bbox)
}
progress_bar.set_postfix(disp_dict)
progress_bar.update()
if VISUAL:
fig, ax = plt.subplots(figsize=(10, 10))
inputs_plt = inputs.detach().cpu().numpy()
#plt.axes(facecolor='silver')
plt.axis([-35, 35, 0, 70])
plt.scatter(inputs_plt[:, 0],
inputs_plt[:, 2],
s=15,
c=inputs_plt[:, 1],
edgecolor='none',
cmap=plt.get_cmap('Blues'),
alpha=1,
marker='.',
vmin=-1,
vmax=2)
pred_boxes3d_numpy = pred_boxes3d[0].detach().cpu().numpy()
pred_boxes3d_corner = kitti_utils.boxes3d_to_corners3d(
pred_boxes3d_numpy, rotate=True)
for o in range(pred_boxes3d_corner.shape[0]):
print_box_corner = pred_boxes3d_corner[o]
x1, x2, x3, x4 = print_box_corner[0:4, 0]
z1, z2, z3, z4 = print_box_corner[0:4, 2]
polygon = np.zeros([5, 2], dtype=np.float32)
polygon[0, 0] = x1
polygon[1, 0] = x2
polygon[2, 0] = x3
polygon[3, 0] = x4
polygon[4, 0] = x1
polygon[0, 1] = z1
polygon[1, 1] = z2
polygon[2, 1] = z3
polygon[3, 1] = z4
polygon[4, 1] = z1
line1 = [(x1, z1), (x2, z2)]
line2 = [(x2, z2), (x3, z3)]
line3 = [(x3, z3), (x4, z4)]
line4 = [(x4, z4), (x1, z1)]
(line1_xs, line1_ys) = zip(*line1)
(line2_xs, line2_ys) = zip(*line2)
(line3_xs, line3_ys) = zip(*line3)
(line4_xs, line4_ys) = zip(*line4)
ax.add_line(
Line2D(line1_xs, line1_ys, linewidth=1, color='green'))
ax.add_line(
Line2D(line2_xs, line2_ys, linewidth=1, color='red'))
ax.add_line(
Line2D(line3_xs, line3_ys, linewidth=1, color='red'))
ax.add_line(
Line2D(line4_xs, line4_ys, linewidth=1, color='red'))
# gt visualize
if args.test == False and data['gt_boxes3d'].shape[1] > 0:
gt_boxes3d_corner = kitti_utils.boxes3d_to_corners3d(
data['gt_boxes3d'].reshape(-1, 7), rotate=True)
for o in range(gt_boxes3d_corner.shape[0]):
print_box_corner = gt_boxes3d_corner[o]
x1, x2, x3, x4 = print_box_corner[0:4, 0]
z1, z2, z3, z4 = print_box_corner[0:4, 2]
polygon = np.zeros([5, 2], dtype=np.float32)
polygon[0, 0] = x1
polygon[1, 0] = x2
polygon[2, 0] = x3
polygon[3, 0] = x4
polygon[4, 0] = x1
polygon[0, 1] = z1
polygon[1, 1] = z2
polygon[2, 1] = z3
polygon[3, 1] = z4
polygon[4, 1] = z1
line1 = [(x1, z1), (x2, z2)]
line2 = [(x2, z2), (x3, z3)]
line3 = [(x3, z3), (x4, z4)]
line4 = [(x4, z4), (x1, z1)]
(line1_xs, line1_ys) = zip(*line1)
(line2_xs, line2_ys) = zip(*line2)
(line3_xs, line3_ys) = zip(*line3)
(line4_xs, line4_ys) = zip(*line4)
ax.add_line(
Line2D(line1_xs, line1_ys, linewidth=1,
color='yellow'))
ax.add_line(
Line2D(line2_xs, line2_ys, linewidth=1,
color='purple'))
ax.add_line(
Line2D(line3_xs, line3_ys, linewidth=1,
color='purple'))
ax.add_line(
Line2D(line4_xs, line4_ys, linewidth=1,
color='purple'))
plt.savefig('../visual/rcnn.jpg')
# scores thresh
inds = (norm_rcnn_score > cfg.RCNN.SCORE_THRESH) & (
norm_ioun_score > cfg.IOUN.SCORE_THRESH)
#inds = (norm_ioun_score > cfg.IOUN.SCORE_THRESH)
for k in range(1):
cur_inds = inds[k].view(-1)
if cur_inds.sum() == 0:
continue
pred_boxes3d_selected = pred_boxes3d[k, cur_inds]
norm_iou_scores_selected = norm_ioun_score[k, cur_inds]
raw_rcnn_score_selected = raw_rcnn_score[k, cur_inds]
#traditional nms
# NMS thresh rotated nms
# boxes_bev_selected = kitti_utils.boxes3d_to_bev_torch(pred_boxes3d_selected)
# #score NMS
# # boxes_bev_selected[:,-1] += np.pi/2
# keep_idx = iou3d_utils.nms_normal_gpu(boxes_bev_selected, norm_iou_scores_selected, cfg.RCNN.NMS_THRESH).view(-1)
# pred_boxes3d_selected = pred_boxes3d_selected[keep_idx]
# norm_iou_scores_selected = norm_iou_scores_selected[keep_idx]
# raw_rcnn_score_selected = raw_rcnn_score_selected[keep_idx]
#self NMS
sort_boxes = torch.argsort(-norm_iou_scores_selected.view(-1))
pred_boxes3d_selected = pred_boxes3d_selected[sort_boxes]
norm_iou_scores_selected = norm_iou_scores_selected[sort_boxes]
if pred_boxes3d_selected.shape[0] > 1:
keep_id = [0]
iou2d, iou3d = iou3d_utils.boxes_iou3d_gpu(
pred_boxes3d_selected, pred_boxes3d_selected)
for i in range(1, pred_boxes3d_selected.shape[0]):
# if torch.min(prop_prop_distance[:i, i], dim=-1)[0] > 0.3:
if torch.max(iou2d[keep_id, i], dim=-1)[0] < 0.01:
keep_id.append(i)
pred_boxes3d_selected = pred_boxes3d_selected[keep_id]
norm_iou_scores_selected = norm_iou_scores_selected[keep_id]
else:
pred_boxes3d_selected = pred_boxes3d_selected
norm_iou_scores_selected = norm_iou_scores_selected
pred_boxes3d_selected, norm_iou_scores_selected = pred_boxes3d_selected.cpu(
).numpy(), norm_iou_scores_selected.cpu().numpy()
cur_sample_id = sample_id
calib = dataset.get_calib(cur_sample_id)
final_total += pred_boxes3d_selected.shape[0]
image_shape = dataset.get_image_shape(cur_sample_id)
save_kitti_format(cur_sample_id, calib, pred_boxes3d_selected,
final_output_dir, norm_iou_scores_selected,
image_shape)
if VISUAL:
fig, ax = plt.subplots(figsize=(10, 10))
inputs_plt = inputs.detach().cpu().numpy()
# plt.axes(facecolor='silver')
plt.axis([-35, 35, 0, 70])
plt.scatter(inputs_plt[:, 0],
inputs_plt[:, 2],
s=15,
c=inputs_plt[:, 1],
edgecolor='none',
cmap=plt.get_cmap('Blues'),
alpha=1,
marker='.',
vmin=-1,
vmax=2)
pred_boxes3d_numpy = pred_boxes3d_selected
pred_boxes3d_corner = kitti_utils.boxes3d_to_corners3d(
pred_boxes3d_numpy, rotate=True)
for o in range(pred_boxes3d_corner.shape[0]):
print_box_corner = pred_boxes3d_corner[o]
x1, x2, x3, x4 = print_box_corner[0:4, 0]
z1, z2, z3, z4 = print_box_corner[0:4, 2]
polygon = np.zeros([5, 2], dtype=np.float32)
polygon[0, 0] = x1
polygon[1, 0] = x2
polygon[2, 0] = x3
polygon[3, 0] = x4
polygon[4, 0] = x1
polygon[0, 1] = z1
polygon[1, 1] = z2
polygon[2, 1] = z3
polygon[3, 1] = z4
polygon[4, 1] = z1
line1 = [(x1, z1), (x2, z2)]
line2 = [(x2, z2), (x3, z3)]
line3 = [(x3, z3), (x4, z4)]
line4 = [(x4, z4), (x1, z1)]
(line1_xs, line1_ys) = zip(*line1)
(line2_xs, line2_ys) = zip(*line2)
(line3_xs, line3_ys) = zip(*line3)
(line4_xs, line4_ys) = zip(*line4)
ax.add_line(
Line2D(line1_xs, line1_ys, linewidth=1, color='green'))
ax.add_line(
Line2D(line2_xs, line2_ys, linewidth=1, color='red'))
ax.add_line(
Line2D(line3_xs, line3_ys, linewidth=1, color='red'))
ax.add_line(
Line2D(line4_xs, line4_ys, linewidth=1, color='red'))
# gt visualize
if args.test == False and data['gt_boxes3d'].shape[1] > 0:
gt_boxes3d_corner = kitti_utils.boxes3d_to_corners3d(
data['gt_boxes3d'].reshape(-1, 7), rotate=True)
for o in range(gt_boxes3d_corner.shape[0]):
print_box_corner = gt_boxes3d_corner[o]
x1, x2, x3, x4 = print_box_corner[0:4, 0]
z1, z2, z3, z4 = print_box_corner[0:4, 2]
polygon = np.zeros([5, 2], dtype=np.float32)
polygon[0, 0] = x1
polygon[1, 0] = x2
polygon[2, 0] = x3
polygon[3, 0] = x4
polygon[4, 0] = x1
polygon[0, 1] = z1
polygon[1, 1] = z2
polygon[2, 1] = z3
polygon[3, 1] = z4
polygon[4, 1] = z1
line1 = [(x1, z1), (x2, z2)]
line2 = [(x2, z2), (x3, z3)]
line3 = [(x3, z3), (x4, z4)]
line4 = [(x4, z4), (x1, z1)]
(line1_xs, line1_ys) = zip(*line1)
(line2_xs, line2_ys) = zip(*line2)
(line3_xs, line3_ys) = zip(*line3)
(line4_xs, line4_ys) = zip(*line4)
ax.add_line(
Line2D(line1_xs,
line1_ys,
linewidth=1,
color='yellow'))
ax.add_line(
Line2D(line2_xs,
line2_ys,
linewidth=1,
color='purple'))
ax.add_line(
Line2D(line3_xs,
line3_ys,
linewidth=1,
color='purple'))
ax.add_line(
Line2D(line4_xs,
line4_ys,
linewidth=1,
color='purple'))
plt.savefig('../visual/ioun.jpg')
progress_bar.close()
# dump empty files
split_file = os.path.join(dataset.imageset_dir, '..', 'ImageSets',
dataset.split + '.txt')
split_file = os.path.abspath(split_file)
image_idx_list = [x.strip() for x in open(split_file).readlines()]
empty_cnt = 0
for k in range(image_idx_list.__len__()):
cur_file = os.path.join(final_output_dir, '%s.txt' % image_idx_list[k])
if not os.path.exists(cur_file):
with open(cur_file, 'w') as temp_f:
pass
empty_cnt += 1
logger.info('empty_cnt=%d: dump empty file %s' %
(empty_cnt, cur_file))
ret_dict = {'empty_cnt': empty_cnt}
if not args.eval_all:
logger.info(
'-------------------performance of epoch %s---------------------' %
epoch_id)
logger.info(str(datetime.now()))
avg_rpn_iou = (total_rpn_iou / max(cnt, 1.0))
avg_cls_acc = (total_cls_acc / max(cnt, 1.0))
avg_cls_acc_refined = (total_cls_acc_refined / max(cnt, 1.0))
avg_det_num = (final_total / max(len(dataset), 1.0))
logger.info('final average detections: %.3f' % avg_det_num)
logger.info('final average rpn_iou refined: %.3f' % avg_rpn_iou)
logger.info('final average cls acc: %.3f' % avg_cls_acc)
logger.info('final average cls acc refined: %.3f' %
avg_cls_acc_refined)
ret_dict['rpn_iou'] = avg_rpn_iou
ret_dict['rcnn_cls_acc'] = avg_cls_acc
ret_dict['rcnn_cls_acc_refined'] = avg_cls_acc_refined
ret_dict['rcnn_avg_num'] = avg_det_num
for idx, thresh in enumerate(thresh_list):
cur_recall = total_recalled_bbox_list[idx] / max(
total_gt_bbox, 1.0)
logger.info('total bbox recall(thresh=%.3f): %d / %d = %f' %
(thresh, total_recalled_bbox_list[idx], total_gt_bbox,
cur_recall))
ret_dict['rcnn_recall(thresh=%.2f)' % thresh] = cur_recall
if thresh == 0.7:
recall = cur_recall
if cfg.TEST.SPLIT != 'test':
logger.info('Averate Precision:')
name_to_class = {'Car': 0, 'Pedestrian': 1, 'Cyclist': 2}
ap_result_str, ap_dict = kitti_evaluate(
dataset.label_dir,
final_output_dir,
label_split_file=split_file,
current_class=name_to_class[cfg.CLASSES])
if not args.eval_all:
logger.info(ap_result_str)
ret_dict.update(ap_dict)
logger.info('result is saved to: %s' % result_dir)
precision = ap_dict['Car_3d_easy'] + ap_dict['Car_3d_moderate'] + ap_dict[
'Car_3d_hard']
recall = total_recalled_bbox_list[3] / max(total_gt_bbox, 1.0)
F2_score = 0
return precision, recall, F2_score
