def visualize_one_sample(self, old_points, expand_points, gt_box_3d,
prop_box_3d, box2d_center_rect):
import mayavi.mlab as mlab
from mayavi_utils import draw_lidar, draw_lidar_simple, draw_gt_boxes3d
# fig = draw_lidar(pc_in_prop_box, pts_color=(1,1,1))
fig = draw_lidar(expand_points[:, :3], pts_color=(1, 1, 1))
fig = draw_lidar(old_points[:, :3], pts_color=(0, 1, 0), fig=fig)
fig = draw_gt_boxes3d([gt_box_3d], fig, color=(1, 0, 0))
fig = draw_gt_boxes3d([prop_box_3d],
fig,
draw_text=False,
color=(1, 1, 1))
# roi_feature_map
# roi_features_size = 7 * 7 * 32
# img_roi_features = prop.roi_features[0:roi_features_size].reshape((7, 7, -1))
# bev_roi_features = prop.roi_features[roi_features_size:].reshape((7, 7, -1))
# img_roi_features = np.amax(img_roi_features, axis=-1)
# bev_roi_features = np.amax(bev_roi_features, axis=-1)
# fig1 = mlab.figure(figure=None, bgcolor=(0,0,0),
# fgcolor=None, engine=None, size=(500, 500))
# fig2 = mlab.figure(figure=None, bgcolor=(0,0,0),
# fgcolor=None, engine=None, size=(500, 500))
# mlab.imshow(img_roi_features, colormap='gist_earth', name='img_roi_features', figure=fig1)
# mlab.imshow(bev_roi_features, colormap='gist_earth', name='bev_roi_features', figure=fig2)
mlab.plot3d([0, box2d_center_rect[0][0]], [0, box2d_center_rect[0][1]],
[0, box2d_center_rect[0][2]],
color=(1, 1, 1),
tube_radius=None,
figure=fig)
raw_input()
def visualize_proposals(self, pc_rect, prop_boxes, neg_boxes, gt_boxes, pc_seg=None):
import mayavi.mlab as mlab
from mayavi_utils import draw_lidar, draw_gt_boxes3d
fig = draw_lidar(pc_rect)
if pc_seg:
fig = draw_lidar(pc_rect[pc_seg==1], fig=fig, pts_color=(1, 1, 1))
fig = draw_gt_boxes3d(prop_boxes, fig, draw_text=False, color=(1, 0, 0))
fig = draw_gt_boxes3d(neg_boxes, fig, draw_text=False, color=(0, 1, 0))
fig = draw_gt_boxes3d(gt_boxes, fig, draw_text=False, color=(1, 1, 1))
raw_input()
def viz_frame(self, pc_rect, mask, gt_boxes):
import mayavi.mlab as mlab
from mayavi_utils import draw_lidar, draw_lidar_simple, draw_gt_boxes3d
fig = draw_lidar(pc_rect)
fig = draw_lidar(pc_rect[mask == 1], fig=fig, pts_color=(1, 1, 1))
fig = draw_gt_boxes3d(gt_boxes, fig, draw_text=False, color=(1, 1, 1))
raw_input()
def visualize(dataset,
frame_id,
prediction,
seg_mask=None,
show_3d=False,
output_dir=None):
fig_size = (12.42, 3.75)
is_video = type(dataset).__name__ == 'kitti_object_video'
# pred_fig, pred_2d_axes, pred_3d_axes = \
# vis_utils.visualization(dataset.image_dir,
# int(frame_id),
# display=False,
# fig_size=fig_size)
pred_fig, pred_3d_axes = vis_utils.visualize_single_plot(dataset.image_dir,
int(frame_id),
is_video,
flipped=False,
display=False,
fig_size=fig_size)
calib = dataset.get_calibration(frame_id) # 3 by 4 matrix
# 2d visualization
# draw groundtruth
# labels = dataset.get_label_objects(frame_id)
# labels = filter(lambda obj: obj.type in type_whitelist and obj.difficulty in difficulties_whitelist, labels)
# draw_boxes(labels, calib, pred_2d_axes)
# draw prediction on second image
pred_corners = draw_boxes(prediction, calib, pred_3d_axes)
if output_dir:
filename = os.path.join(output_dir,
'result_2d_image/%06d.png' % frame_id)
plt.savefig(filename)
plt.close(pred_fig)
else:
plt.show()
raw_input()
if seg_mask is not None:
img = dataset.get_image(frame_id)
img = cv2.resize(img, (1200, 360))
seg_img = add_mask_to_img(img, seg_mask)
filename = os.path.join(output_dir,
'result_seg_image/%06d.png' % frame_id)
cv2.imwrite(filename, seg_img)
if show_3d:
# 3d visualization
pc_velo = dataset.get_lidar(frame_id)
image = dataset.get_image(frame_id)
img_height, img_width = image.shape[0:2]
_, _, img_fov_inds = get_lidar_in_image_fov(pc_velo[:, 0:3], calib, 0,
0, img_width, img_height,
True)
pc_velo = pc_velo[img_fov_inds, :]
boxes3d_velo = []
mask = np.zeros((len(pc_velo), ))
for corners in pred_corners:
pts_velo = calib.project_rect_to_velo(corners)
boxes3d_velo.append(pts_velo)
_, obj_mask = extract_pc_in_box3d(pc_velo, pts_velo)
mask = np.logical_or(mask, obj_mask)
fig = draw_lidar(pc_velo, pts_color=(1, 1, 1))
fig = draw_lidar(pc_velo[mask == 1], fig=fig, pts_color=(1, 0, 0))
fig = draw_gt_boxes3d(boxes3d_velo,
fig,
draw_text=False,
color=(1, 1, 1))
# raw_input()
if output_dir:
filename = os.path.join(output_dir,
'result_3d_image/%06d.png' % frame_id)
mlab.savefig(filename, figure=fig)