def render_points(points):
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
fig = utils.draw_lidar(points, fig=fig, color_by_intensity=True)
mlab.show()
def render(points, img):
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
fig = utils.draw_lidar(points, fig=fig)
cv2.imshow("image", img)
mlab.show()
def show_predicted_results(sample_id_in_val=None,draw_text=True):
if show_predicted_results is None:
sample_id_in_val = cfg_vis.val_idx
pred_path = cfg_vis.predict_path
val_split_dir = cfg_vis.val_dataset
val_sample_idx = [x.strip() for x in open(val_split_dir).readlines()]
gt_lidar_dir = os.path.join(cfg_vis.kitti_gt_dir, "lidar_gt_list.pth")
with open(gt_lidar_dir, "rb") as f:
gt_lidar_list_with_class = pickle.load(f)
gt_lidar_list = gt_lidar_list_with_class["gt_lidar_box_list"]
gt_class_names_list = gt_lidar_list_with_class["gt_class_names"]
sample_idx_str = val_sample_idx[sample_id_in_val]
sample_idx = int(sample_idx_str)
points_path = os.path.join(cfg_vis.KITTI_DIR, "velodyne", sample_idx_str + ".bin")
point_xyzi = np.fromfile(points_path, dtype=np.float32).reshape(-1, 4)
image_path = os.path.join(cfg_vis.KITTI_DIR, "image_2", sample_idx_str + ".png")
image = cv2.imread(image_path)
with open(pred_path, "rb") as f:
pred_dicts = pickle.load(f)
det_anno = pred_dicts[sample_id_in_val]
det_boxes = det_anno['boxes_lidar']
det_class_name = list(det_anno["name"])
gt_boxes = gt_lidar_list[sample_id_in_val]
gt_classes = gt_class_names_list[sample_id_in_val]
gt_boxes_corners = boxes3d_to_corners3d_lidar(gt_boxes[:, :7])
det_boxes_corners = boxes3d_to_corners3d_lidar(det_boxes[:, :7])
fig = mlab.figure(
figure=None, bgcolor=(0, 0, 0), fgcolor=None, engine=None, size=(1600, 1000)
)
fig = utils.draw_gt_boxes3d(gt_boxes_corners, color=(1, 0, 0), fig=fig)
if draw_text:
fig = utils.draw_gt_boxes3d(det_boxes_corners, color=(0, 1, 0), label=det_class_name, draw_text=True, fig=fig)
else:
fig = utils.draw_gt_boxes3d(det_boxes_corners, color=(0, 1, 0), fig=fig)
fig = utils.draw_lidar(point_xyzi[:, 0:3], fig=fig)
cv2.imshow("image", image)
mlab.show()
def main1():
data_root_path = cfg_vis.DATA_DIR
lidar_points = get_lidar("000211", data_root_path, "training")
gt_lidar_box = get_labels("000211", data_root_path, "training")
# features= model(lidar_points)
plot_str = "fps_f"
if plot_str == "fps_f":
seg_features = points_in_boxes_cpu(
torch.from_numpy(lidar_points[:, :3]),
torch.from_numpy(gt_lidar_box[:, :7])).numpy()
seg_features = seg_features.sum(axis=0) # segment labels 0 or 1
seg_features = np.where(seg_features > 0, 1, 0)
seg_features = seg_features[np.newaxis, ..., np.newaxis]
lidar_points = lidar_points[np.newaxis, ...]
seg_features = torch.from_numpy(seg_features).type(torch.int32).cuda()
lidar_points = torch.from_numpy(lidar_points).cuda()
keypoints = fps_f(lidar_points, seg_features, 1024)
elif plot_str == "fps_d":
lidar_points = lidar_points[np.newaxis, ...]
keypoints = fps_d(lidar_points, 1024)
else:
assert NotImplementedError
points_in_boxes, points_out_boxes = utils.points_indices(
keypoints, gt_lidar_box)
points_size = np.array([3]).repeat(points_in_boxes.shape[0], 0)
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
fig = utils.draw_lidar(points_out_boxes, fig=fig)
gt_boxes_corners = boxes3d_to_corners3d_lidar(gt_lidar_box[:, :7])
fig = utils.draw_gt_boxes3d(gt_boxes_corners, color=(1, 0, 0), fig=fig)
mlab.points3d(points_in_boxes[:, 0],
points_in_boxes[:, 1],
points_in_boxes[:, 2],
points_size,
color=(1, 0, 0),
mode="sphere",
scale_factor=0.1,
figure=fig)
mlab.show()
def show_box_with_anno(ann_dir, bin_dir, image_dir):
anno_list = glob.glob(os.path.join(ann_dir, "*.txt"))
image_list = glob.glob(os.path.join(image_dir, "*.png"))
image_list.sort()
anno_list.sort()
bin_list = glob.glob(os.path.join(bin_dir, "*.bin"))
bin_list.sort()
for anno_path, i in zip(anno_list, range(len(bin_list))):
id = anno_path.split("/")[-1]
id = id.split(".")[0]
print("frame id:%s" % id)
if id != "000543":
continue
image_path = os.path.join(image_dir, id + ".png")
bin_path = os.path.join(bin_dir, id + ".bin")
objects = get_objects_from_label(anno_path)
gt_boxes = get_info(objects)
print("gt_boxes:", gt_boxes)
gt_boxes_corners, temp_corners = center2corner_leishen(gt_boxes[:, :7])
points = read_bin(bin_path)
pts = np.unique(points, axis=0).reshape(-1, 4)
img = cv2.imread(image_path)
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
fig = utils.draw_lidar(pts, fig=fig, datatype="leishen")
fig = utils.draw_gt_boxes3d(gt_boxes_corners,
color=(1, 0, 0),
fig=fig,
gt_boxes_center=gt_boxes)
cv2.imshow("image", img)
mlab.show()
print("done")
def show_single_frame():
data_dir = "/media/liang/Elements/rosbag/leishen_e70_32/dataset_image_pcd"
bin_dir = os.path.join(data_dir, "bin")
id = 552
anno_path = os.path.join(data_dir, "annotation", str(id).zfill(6) + ".txt")
bin_path = os.path.join(bin_dir, str(id).zfill(6) + ".bin")
objects = get_objects_from_label(anno_path)
gt_boxes = get_info(objects)
print("gt_boxes:", gt_boxes)
gt_boxes_corners, temp_corners = center2corner_leishen(gt_boxes[:, :7])
points = read_bin(bin_path)
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
fig = utils.draw_lidar(points, fig=fig, datatype="leishen")
fig = utils.draw_gt_boxes3d(gt_boxes_corners,
color=(1, 0, 0),
fig=fig,
gt_boxes_center=gt_boxes)
mlab.show()
def show_frame(id=1,
show_augmatation=False,
show_predict=False,
eval_save_pth=False):
image_dir = os.path.join(DATA_DIR, "image", "sorted")
image_path = os.path.join(image_dir, str(id).zfill(6) + ".png")
image_data = cv2.imread(image_path)
if show_augmatation:
anno_path = os.path.join(DATA_DIR, "augmentation",
str(id).zfill(6) + ".txt")
bin_dir = os.path.join(DATA_DIR, "augmentation")
bin_path = os.path.join(bin_dir, str(id).zfill(6) + ".bin")
else:
anno_path = os.path.join(DATA_DIR, "annotation",
str(id).zfill(6) + ".txt")
bin_dir = os.path.join(DATA_DIR, "bin")
bin_path = os.path.join(bin_dir, str(id).zfill(6) + ".bin")
if show_predict and not show_augmatation:
# "/for_ubuntu502/PVRCNN-V1.1/output/single_stage_model/leishen_1.1/eval/epoch_80/final_result/data"
pred_path = os.path.join(OUTPUT_DIR, "single_stage_model",
"leishen_1.3", "eval", "epoch_80",
"final_result", "data",
str(id).zfill(6) + ".txt")
predict_objects = get_objects_from_label(pred_path)
det_boxes = get_info(predict_objects, drop_bus=True)
if eval_save_pth:
result_dir = os.path.join(CODE_DIR, "output", "single_stage_model",
"leishen_1.3", "eval", "epoch_80")
reuslt_file_path = os.path.join(result_dir, "result.pkl")
with open(reuslt_file_path, "rb") as f:
det_annos = pickle.load(f)
for sample in det_annos:
sample["frame_id"] == str(id).zfill(6)
break
det_boxes_ = sample['boxes_lidar']
objects = get_objects_from_label(anno_path)
gt_boxes = get_info(objects)
if show_predict:
if len(gt_boxes) > 0:
values = iou3d_nms_utils.boxes_iou3d_gpu(
torch.from_numpy(gt_boxes[:, :7].astype(np.float32)).cuda(),
torch.from_numpy(det_boxes[:, :7].astype(
np.float32)).cuda()).cpu().numpy()
# gt_boxes = gt_boxes[6:7,:]
# det_boxes = det_boxes[8:9,:]
if len(gt_boxes) > 0:
gt_boxes_corners, temp_corners = center2corner_leishen(gt_boxes[:, :7])
if show_predict:
if len(det_boxes) > 0:
det_boxes_corners, det_temp_corners = center2corner_leishen(
det_boxes[:, :7])
points = np.fromfile(bin_path, dtype=np.float32).reshape(-1, 4)
# points = anno_utils.remove_points_in_boxes3d_v0(points,gt_boxes[0:1,:7])
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
# mlab.points3d(
# points[:, 0],
# points[:, 1],
# points[:, 2],
# color=(1, 1, 1),
# mode="point",
# colormap="gnuplot",
# scale_factor=0.3,
# figure=fig,
# )
fig = utils.draw_lidar(points,
fig=fig,
datatype="leishen",
draw_axis=True,
draw_square_region=False,
draw_fov=False)
if len(gt_boxes) > 0:
fig = utils.draw_gt_boxes3d(gt_boxes_corners,
color=(1, 0, 0),
fig=fig,
gt_boxes_center=gt_boxes)
if show_predict:
fig = utils.draw_gt_boxes3d(det_boxes_corners,
color=(0, 1, 0),
fig=fig,
gt_boxes_center=gt_boxes)
cv2.imshow("image", image_data)
mlab.show()
if i != cfg_vis.val_idx:
continue
print(i)
id = val_sample_idx[i]
point_file = id + ".bin"
lidar_path = os.path.join(cfg_vis.KITTI_DIR, "velodyne",
point_file)
image_name = id + ".png"
image_path = os.path.join(cfg_vis.KITTI_DIR, "image_2", image_name)
points = np.fromfile(lidar_path, dtype=np.float32).reshape(-1, 4)
fig = mlab.figure(figure=None,
bgcolor=(0, 0, 0),
fgcolor=None,
engine=None,
size=(1600, 1000))
fig = utils.draw_lidar(points, fig=fig)
image_file = cv2.imread(image_path)
cv2.imshow("image", image_file)
mlab.show()
if cfg_vis.SHOW == "center_points":
split = "training"
point_file = cfg_vis.SAMPLE_INDEX + ".bin"
lidar_path = os.path.join(cfg_vis.KITTI_DIR, point_file)
points = np.fromfile(lidar_path, dtype=np.float32).reshape(-1, 4)
gt_lidar_box = get_labels(cfg_vis.SAMPLE_INDEX, cfg_vis.DATA_DIR,
split)
gt_boxes_corners = boxes3d_to_corners3d_lidar(gt_lidar_box[:, :7])
points_in_boxes, points_out_boxes = utils.points_indices(
points, gt_lidar_box)
