def build_graph():
input_image = tf.placeholder(tf.float32, shape=[None, None, None, 3], name='image')
input_label = tf.placeholder(tf.int32, shape=[None, 1], name='labels')
input_bbox_targets = tf.placeholder(tf.float32, shape=[None, 4], name='gt_boxes')
input_bbox_inside_weights = tf.placeholder(tf.float32, shape=[None, 4], name="bbox_inside_weights")
input_bbox_outside_weights = tf.placeholder(tf.float32, shape=[None, 4], name='bbox_outside_weights')
learing_rate = tf.placeholder(tf.float32)
cnn_layers = ResNet()
rnn_layers = Lstm()
x = input_image
x = cnn_layers.build(x, True)
x = rnn_layers.build(x, 512,128, 512)
########################
# rpn cls score
########################
y = fc_layer(x, 512, 10 * 2, "fc_rpn_cls")
dims = tf.shape(y)
cls_prob = tf.reshape(tf.nn.softmax(tf.reshape(y, [-1, 2])),
[dims[0], dims[1], -1, 2])
#########################
# rpn bbox pred
#########################
box_pred = fc_layer(x, 512, 10 * 4, "fc_rpn_pred")
loss_layer = Loss()
output_loss ,cls_loss,box_loss = loss_layer.build(y,box_pred,input_label,input_bbox_targets,input_bbox_inside_weights,input_bbox_outside_weights)
train_step = tf.train.AdamOptimizer(learing_rate).minimize(output_loss)
return [train_step,output_loss,learing_rate,input_image,
input_label,input_bbox_targets,input_bbox_inside_weights,input_bbox_outside_weights,cls_prob,box_pred,cls_loss,box_loss]
def build_graph():
input_image = tf.placeholder(tf.float32,
shape=[None, None, None, 3],
name='image')
cnn_layers = ResNet()
rnn_layers = Lstm()
x = input_image
x = cnn_layers.build(x, True)
x = rnn_layers.build(x, 512, 128, 512)
########################
# rpn cls score
########################
y = fc_layer(x, 512, 10 * 2, "fc_rpn_cls")
dims = tf.shape(y)
cls_prob = tf.reshape(tf.nn.softmax(tf.reshape(y, [-1, 2])),
[dims[0], dims[1], -1, 2])
#########################
# rpn bbox pred
#########################
box_pred = fc_layer(x, 512, 10 * 4, "fc_rpn_pred")
return [input_image, cls_prob, box_pred]
class Inference:
def __init__(self):
self.net = ResNet(inference_mode=True)
self.net.build(input_shape=(None, *IMAGE_SIZE, CHANNELS))
self.net.load_weights(PATH_TO_MODEL_WEIGHT)
self.net.summary()
self.dataset = DataController(PATH_TO_TEST_DATA, True, draw_cloud=True)
self.data_generator = self.dataset.get_eval_data_generator(
shuffle=True)
def __call__(self):
for images, gt_labels in self.data_generator:
scores = tf.nn.softmax(self.net(images)).numpy()
for i in range(INFER_BATCH_SIZE):
label = int(scores[i][0] < scores[i][1])
yield 'ALERT' if label else 'PASS'
def main():
if VISUALIZATION:
window_manager = VisualManager('eval')
dataset = DataController(PATH_TO_TEST_DATA, inference_mode=True)
measure = YieldMeasure()
data_generator = dataset.get_eval_data_generator(shuffle=True)
net = ResNet(inference_mode=True)
net.build(input_shape=(None, *IMAGE_SIZE, CHANNELS))
net.load_weights(PATH_TO_MODEL_WEIGHT)
net.summary()
inf_time = 0
inf_num = 0
for images, gt_labels in data_generator:
gt_labels = gt_labels.numpy()
t0 = time()
pred_labels = tf.nn.softmax(
net(images)).numpy().argmax(axis=1).astype('int32')
inf_time += (time() - t0)
inf_num += 1
measure.append(gt_labels, pred_labels)
if VISUALIZATION:
visual_batch = []
for i in range(INFER_BATCH_SIZE):
frame = (images[i] * 255).numpy().astype('uint8')
window_manager.draw_annotations(frame, pred_labels[i],
gt_labels[i],
dataset.invert_class_dict)
visual_batch.append(frame)
window_manager.desktop_show(concatenate(visual_batch, axis=1), 0)
measure.show_rating()
print(f"Median inference time: {(inf_time / inf_num) * 1000:.1f} ms")