def inference():
config_name = 'config.json'
config, _, _, _ = load_config(config_name)
if torch.cuda.is_available():
device = 'cuda'
net = LDOPC(config['use_bn']).cuda()
else:
device = 'cpu'
net = LDOPC(config['use_bn']).cpu()
net.load_state_dict(
torch.load(get_model_name(config['name']), map_location=device))
net.set_decode(True)
loader, _ = get_data_loader(batch_size=1,
use_npy=config['use_npy'],
frame_range=config['frame_range'])
net.eval()
image_id = 25
threshold = config['cls_threshold']
with torch.no_grad():
pc_feature, label_map = loader.dataset[image_id]
pc_feature = pc_feature.to(device)
label_map = label_map.to(device)
label_map_unnorm, label_list = loader.dataset.get_label(image_id)
t_start = time.time()
pred = net(pc_feature.unsqueeze(0)).squeeze_(0)
print("Forward pass time", time.time() - t_start)
cls_pred = pred[..., 0]
activation = cls_pred > threshold
num_boxes = int(activation.sum())
if num_boxes == 0:
print("No bounding box found")
return
corners = torch.zeros((num_boxes, 8))
for i in range(1, 9):
corners[:, i - 1] = torch.masked_select(pred[..., i], activation)
corners = corners.view(-1, 4, 2).numpy()
scores = (torch.masked_select(pred[..., 0], activation)).cpu().numpy()
t_start = time.time()
selected_ids = non_max_suppression(corners, scores,
config['nms_iou_threshold'])
corners = corners[selected_ids]
scores = scores[selected_ids]
print("Non max suppression time:", time.time() - t_start)
pc_feature = pc_feature.cpu().numpy()
plot_bev(pc_feature, label_list, window_name='GT')
plot_bev(pc_feature, corners, window_name='Prediction')
plot_label_map(cls_pred.cpu().numpy())
cv2.waitKey(0)
def experiment(config_name, device):
config, _, _, _ = load_config(config_name)
net, criterion = build_model(config, device, train=False)
net.load_state_dict(
torch.load(get_model_name(config['name']), map_location=device))
net.set_decode(True)
loader, _ = get_data_loader(batch_size=1,
use_npy=config['use_npy'],
frame_range=config['frame_range'])
net.eval()
image_id = 25
threshold = config['cls_threshold']
with torch.no_grad():
input, label_map = loader.dataset[image_id]
input = input.to(device)
label_map = label_map.to(device)
label_map_unnorm, label_list = loader.dataset.get_label(image_id)
# Forward Pass
t_start = time.time()
pred = net(input.unsqueeze(0)).squeeze_(0)
print("Forward pass time", time.time() - t_start)
# Select all the bounding boxes with classification score above threshold
cls_pred = pred[..., 0]
activation = cls_pred > threshold
# Compute (x, y) of the corners of selected bounding box
num_boxes = int(activation.sum())
if num_boxes == 0:
print("No bounding box found")
return
corners = torch.zeros((num_boxes, 8))
for i in range(1, 9):
corners[:, i - 1] = torch.masked_select(pred[..., i], activation)
corners = corners.view(-1, 4, 2).numpy()
scores = torch.masked_select(pred[..., 0], activation).numpy()
# NMS
t_start = time.time()
selected_ids = non_max_suppression(corners, scores,
config['nms_iou_threshold'])
corners = corners[selected_ids]
scores = scores[selected_ids]
print("Non max suppression time:", time.time() - t_start)
# Visualization
input_np = input.cpu().numpy()
plot_bev(input_np, label_list, window_name='GT')
plot_bev(input_np, corners, window_name='Prediction')
plot_label_map(cls_pred.numpy())
def eval_one(net,
loss_fn,
config,
loader,
image_id,
device,
plot=False,
verbose=False):
input, label_map, image_id = loader.dataset[image_id]
input = input.to(device)
label_map, label_list = loader.dataset.get_label(image_id)
loader.dataset.reg_target_transform(label_map)
label_map = torch.from_numpy(label_map).permute(2, 0,
1).unsqueeze_(0).to(device)
# Forward Pass
t_start = time.time()
pred = net(input.unsqueeze(0))
t_forward = time.time() - t_start
loss, cls_loss, loc_loss = loss_fn(pred, label_map)
pred.squeeze_(0)
cls_pred = pred[0, ...]
if verbose:
print("Forward pass time", t_forward)
# Filter Predictions
t_start = time.time()
corners, scores = filter_pred(config, pred)
t_post = time.time() - t_start
if verbose:
print("Non max suppression time:", t_post)
gt_boxes = np.array(label_list)
gt_match, pred_match, overlaps = compute_matches(gt_boxes,
corners,
scores,
iou_threshold=0.5)
num_gt = len(label_list)
num_pred = len(scores)
input_np = input.cpu().permute(1, 2, 0).numpy()
pred_image = get_bev(input_np, corners)
if plot == True:
# Visualization
plot_bev(input_np, label_list, window_name='GT')
plot_bev(input_np, corners, window_name='Prediction')
plot_label_map(cls_pred.numpy())
return num_gt, num_pred, scores, pred_image, pred_match, loss.item(
), t_forward, t_post
def test0():
k = KITTI()
id = 25
k.load_velo()
tstart = time.time()
scan = k.load_velo_scan(id)
processed_v = k.lidar_preprocess(scan)
label_map, label_list = k.get_label(id)
print('time taken: %gs' % (time.time() - tstart))
plot_bev(processed_v, label_list)
plot_label_map(label_map[:, :, 6])