end_image = (rank + 1) * images_per_process
print("Rank %d, Start %d, End %d" % (rank, start_image, end_image))
# object detection on images for this proces
total_runtime = 0
for _id, _url in enumerate(
tqdm(image_urls[start_image:end_image], desc="Rank %d" % rank)):
path = download_and_resize_image(_url, 640, 480)
img = tf.io.read_file(path)
img = tf.image.decode_jpeg(img, channels=3)
converted_img = tf.image.convert_image_dtype(img, tf.float32)[tf.newaxis,
...]
start_time = time.time()
result = detector(converted_img)
end_time = time.time()
result = {key: value.numpy() for key, value in result.items()}
objects = list(set(result['detection_class_entities'].tolist()))
total_runtime = total_runtime + (end_time - start_time)
# print("Rank %d, image %s, runtime %0.2f"%(rank,path,end_time-start_time))
mlflow_params["objects"] = objects
mlflow_metrics["runtime"] = end_time - start_time
image_with_boxes = draw_boxes(img.numpy(), result["detection_boxes"],
result["detection_class_entities"],
result["detection_scores"])
filename = save_to_file(image_with_boxes)
picture_labels.append(picture_label)
# Select Neurons
print("Selecting Neurons...")
selected = select_neurons(
model, target_layers, image_data, base_vars, config,
activation_cases=base_vars["activation_cases"])
selected_weight.append(selected)
if picture_label == 1:
# Draw Boxes
print("Drawing Boxes")
for activation_case, selected_metric in zip(base_vars["activation_cases"], selected):
vars_ = unpack(base_vars, activation_case, config)
draw_boxes(
patches, selected_metric, vars_["img_dir"],
picture_name, vars_["width"],
vars_["height"], vars_['bounding_box_path'])
# Calculate the Overlap between the selections by the
# Activation-Output Correlation metric and the Precision metric
print("Calculating Overlap")
overlap = analysis.calc_overlap(selected[4], selected[5])
overlaps_weight.append(overlap)
print("Overlap Size: {}".format(overlap))
overlaps.append(overlaps_weight)
# Calculate the accuracy for each selection metric
print("Calculating Accuracies")
accuracies_weight = []
for i in range(len(selected_weight[0])):
X_train, X_test, Y_train, Y_test = get_train_test_sets(
[430, 430]])
''' Example code to generate adv_example/color_filtering.png '''
# img = color_filtering(img)
''' Example code to generate adv_example/perspective.png '''
# img = blur(img, (5,5))
#cv2.imwrite('layers_car/large_6.jpg', img)
#objects = localize_objects('layers_car/large_12.jpg')
ret = detect_safe_search('images/0000.png')
print(ret)
'''
for i in range(0, 2000):
ret = detect_safe_search('images/0000.png')
print(ret)
'''
'''
labels = []
coords = []
image = Image.open("layers_car/large_12.jpg")
for object in objects:
label = object.name + ":" + str(object.score)[:4]
labels.append(label)
top = object.bounding_poly.normalized_vertices[0].y
left = object.bounding_poly.normalized_vertices[0].x
bottom = object.bounding_poly.normalized_vertices[2].y
right = object.bounding_poly.normalized_vertices[2].x
coords.append([top, left, bottom, right])
draw_boxes(image, labels, coords)
image.save("images/out.png", "PNG")
'''