def StartTrain():
loader = load.Loader()
batch = cfg.BATCH_SIZE
if cfg.SPECIFIC_DATA != "":
loader.load_specific_label(cfg.SPECIFIC_DATA)
batch = 1
else:
loader.load_data()
nc = NetworkCreator(batch)
nc.train(loader)
def scale_box_size():
loader = load.Loader()
loader.load_data()
bound_above_2, bound_below_2, _ = GetObjectBounds(2, 0.3, 0.33, 2)
bound_above_4, bound_below_4, _ = GetObjectBounds(2, 0.3, 0.33, 4)
bound_above_8, bound_below_8, _ = GetObjectBounds(2, 0.3, 0.33, 8)
bound_above_16, bound_below_16, _ = GetObjectBounds(2, 0.3, 0.33, 16)
heights_scale_2 = []
heights_scale_4 = []
heights_scale_8 = []
heights_scale_16 = []
for im_model in loader.Data:
for label in im_model.labels:
if label.largest_dim >= bound_below_2 and label.largest_dim <= bound_above_2:
dist = np.abs(label.fbl_y - label.ftl_y)
heights_scale_2.append(dist)
if label.largest_dim >= bound_below_4 and label.largest_dim <= bound_above_4:
dist = np.abs(label.fbl_y - label.ftl_y)
heights_scale_4.append(dist)
if label.largest_dim >= bound_below_8 and label.largest_dim <= bound_above_8:
dist = np.abs(label.fbl_y - label.ftl_y)
heights_scale_8.append(dist)
if label.largest_dim >= bound_below_16 and label.largest_dim <= bound_above_16:
dist = np.abs(label.fbl_y - label.ftl_y)
heights_scale_16.append(dist)
_min = np.mean(heights_scale_2) - 2 * np.std(heights_scale_2)
_max = np.mean(heights_scale_2) + 2 * np.std(heights_scale_2)
print("heights 2 min:", _min, " max:", _max)
_min = 0
_max = 0
_min = np.mean(heights_scale_4) - 2 * np.std(heights_scale_4)
_max = np.mean(heights_scale_4) + 2 * np.std(heights_scale_4)
print("heights 4 min:", _min, " max:", _max)
_min = 0
_max = 0
_min = np.mean(heights_scale_8) - 2 * np.std(heights_scale_8)
_max = np.mean(heights_scale_8) + 2 * np.std(heights_scale_8)
print("heights 8 min:", _min, " max:", _max)
_min = 0
_max = 0
_min = np.mean(heights_scale_16) - 2 * np.std(heights_scale_16)
_max = np.mean(heights_scale_16) + 2 * np.std(heights_scale_16)
print("heights 16 min:", _min, " max:", _max)
def start_test():
if cfg.SPECIFIC_TEST_DATA == "":
print("Test data not specified!!! Check config file")
return
loader = load.Loader()
loader.load_specific_label(cfg.SPECIFIC_TEST_DATA)
loader.prepare_data(1)
new_saver = tf.train.import_meta_graph(cfg.MODEL_PATH + '.meta',
clear_devices=True)
graph = tf.get_default_graph()
input_graph_def = graph.as_graph_def()
with tf.Session() as sess:
new_saver.restore(sess, cfg.MODEL_PATH)
graph = tf.get_default_graph()
image_placeholder = graph.get_tensor_by_name(
"input_image_placeholder:0")
predict_2 = graph.get_tensor_by_name("scale_2/output2/Conv2D:0")
predict_4 = graph.get_tensor_by_name("scale_4/output4/Conv2D:0")
predict_8 = graph.get_tensor_by_name("scale_8/output8/Conv2D:0")
predict_16 = graph.get_tensor_by_name("scale_16/output16/Conv2D:0")
image_batch, label_batch, image_paths, image_names, calib_matrices = loader.get_test_data(
1)
response_maps_2, response_maps_4, response_maps_8, response_maps_16 = sess.run(
[predict_2, predict_4, predict_8, predict_16],
feed_dict={image_placeholder: image_batch})
save_results(response_maps_2, 2)
save_results(response_maps_4, 4)
save_results(response_maps_8, 8)
save_results(response_maps_16, 16)
extract_and_show(response_maps_2, response_maps_4, response_maps_8,
response_maps_16, label_batch, calib_matrices,
image_paths)
data = np.asarray([w_fbl, w_fbr, w_rbl, w_rbr, w_ftl, w_ftr, w_rtl, w_rtr])
transposed = np.squeeze(data.transpose())
data = np.ones((4, 8))
data[0:3, 0:8] = transposed
points = world_space_to_image(data, calib_matrix)
box = BoxModel()
box.fbl = (int(points[0, 0]), int(points[1, 0]))
box.fbr = (int(points[0, 1]), int(points[1, 1]))
box.rbl = (int(points[0, 2]), int(points[1, 2]))
box.rbr = (int(points[0, 3]), int(points[1, 3]))
box.ftl = (int(points[0, 4]), int(points[1, 4]))
box.ftr = (int(points[0, 5]), int(points[1, 5]))
box.rtl = (int(points[0, 6]), int(points[1, 6]))
box.rtr = (int(points[0, 7]), int(points[1, 7]))
image_model.boxes.append(box)
return image_model
if __name__ == "__main__":
loader = load.Loader()
# loader.load_data()
loader.load_specific_label("000046")
image_batch, label_batch, image_paths, calib_matrices = loader.get_test_data(
1)
b_boxes_model = test_projection(label_batch[0][1], calib_matrices)
b_boxes_model.file_name = image_paths[0]
drawer.draw_bounding_boxes(b_boxes_model)
def test_target_map_creation(name):
loader = load.Loader()
loader.load_specific_label(name)
label = tf.placeholder(tf.float32,(None,None,10), name="label")
target = tf.py_func(create_target_response_map, [label[0], 128, 64, 8, 2,0.3, 0.33, 2], [tf.float32])
target2 = tf.reshape(target,(8,64,128))
tmp = tf.py_func(create_target_response_map, [label[0], 64, 32, 8, 2,0.3, 0.33, 4], [tf.float32])
target4 = tf.reshape(tmp,(8,32,64))
tmp = tf.py_func(create_target_response_map, [label[0], 32, 16, 8, 2,0.3, 0.33, 8], [tf.float32])
target8 = tf.reshape(tmp,(8,16,32))
tmp = tf.py_func(create_target_response_map, [label[0], 16, 8, 8, 2,0.3, 0.33, 16], [tf.float32])
target16 = tf.reshape(tmp,(8,8,16))
initial = tf.Variable(tf.zeros_like(target2[0,:, :]),dtype=tf.float32, name="initial")
tmp_initial = initial
condition = tf.greater(target2[0,:, :], tf.constant(0,dtype=tf.float32),name="greater")
weight_factor_array = initial.assign( tf.where(condition, (tmp_initial + 2.0), tmp_initial, name="where_condition"), name="assign" )
init = tf.global_variables_initializer()
errors = [2.0,2.0]
loss_output = tf.placeholder(dtype=tf.float32,shape=(2),name='loss_output_placeholder')
loss_constant = tf.constant(1.0,shape=[2],dtype=tf.float32,name='loss_constant')
loss_output = tf.multiply(errors,loss_constant, name='loss_output_multiply')
image_batch, labels_batch, image_paths,image_names, calib_matrices = loader.get_test_data(1)
# maps = create_target_response_map(labels_batch[0], 128, 64, 8, 2,0.3, 0.33, 4)
# print(maps.shape)
# tmp = h.change_first_x_last_dim(maps)
# cv2.imshow("target 1", maps[0,:,:])
# cv2.imshow("target 2", maps[1,:,:])
# cv2.imshow("target 3", maps[2,:,:])
# cv2.imshow("target 4", maps[3,:,:])
# cv2.imshow("target 5", maps[4,:,:])
# cv2.imshow("target 6", maps[5,:,:])
# cv2.imshow("target 7", maps[6,:,:])
# cv2.imshow("target 8", maps[7,:,:])
# cv2.imshow("target 11", tmp[:,:,0])
# cv2.imshow("target 22", tmp[:,:,1])
# cv2.imshow("target 33", tmp[:,:,2])
# cv2.imshow("target 44", tmp[:,:,3])
# cv2.imshow("target 55", tmp[:,:,4])
# cv2.imshow("target 66", tmp[:,:,5])
# cv2.imshow("target 77", tmp[:,:,6])
# cv2.imshow("target 88", tmp[:,:,7])
# cv2.waitKey(0)
# cv2.waitKey(0)
# tmp = np.reshape(maps,(8,16,8))
# print(tmp)
with tf.Session() as session:
# initialise the variables
session.run(init)
tmp,weight = session.run([target2,weight_factor_array], feed_dict={label: labels_batch})
maps = h.change_first_x_last_dim(tmp)
cv2.imshow("target 2", maps[:,:,0])
cv2.imshow("weights", weight*127)
cv2.waitKey(0)
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\target2.jpg', 255*maps[:,:,0])
resized = cv2.resize(maps[:,:,0], (1242,375), interpolation = cv2.INTER_AREA)
cv2.imshow("resized", resized)
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\resized2.jpg', 255*resized)
# cv2.imshow("target 2", maps[:,:,1])
# cv2.imshow("target 3", maps[:,:,2])
# cv2.imshow("target 4", maps[:,:,3])
# cv2.imshow("target 5", maps[:,:,4])
# cv2.imshow("target 6", maps[:,:,5])
# cv2.imshow("target 7", maps[:,:,6])
# cv2.imshow("target 8", maps[:,:,7])
tmp2 = session.run(target4, feed_dict={label: labels_batch})
maps = h.change_first_x_last_dim(tmp2)
cv2.imshow("target 4", maps[:,:,0])
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\target4.jpg', 255*maps[:,:,0])
resized4 = cv2.resize(maps[:,:,0], (1242,375), interpolation = cv2.INTER_AREA)
cv2.imshow("resized4", resized4)
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\resized4.jpg', 255*resized4)
tmp4 = session.run(target8, feed_dict={label: labels_batch})
maps = h.change_first_x_last_dim(tmp4)
cv2.imshow("target 8", maps[:,:,0])
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\target8.jpg', 255*maps[:,:,0])
resized8 = cv2.resize(maps[:,:,0], (1242,375), interpolation = cv2.INTER_AREA)
cv2.imshow("resized8", resized8)
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\resized8.jpg', 255*resized8)
tmp8 = session.run(target16, feed_dict={label: labels_batch})
maps = h.change_first_x_last_dim(tmp8)
cv2.imshow("target 16", maps[:,:,0])
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\target16.jpg', 255*maps[:,:,0])
resized16 = cv2.resize(maps[:,:,0], (1242,375), interpolation = cv2.INTER_AREA)
cv2.imshow("resized6", resized16)
cv2.imwrite(r'C:\Users\Lukas\Documents\Object detection\test\resized16.jpg', 255*resized16)
cv2.waitKey(0)