def uploaded_file(filename):
# read image file and make prediction
here = os.getcwd()
image_path = os.path.join(here, app.config['UPLOAD_FOLDER'], filename)
image = cv2.imread(image_path)
(class_ids, labels, boxes,
confidences) = yolo_forward(net, LABELS, image, confidence_level=0.3)
# format data for template rendering
# found emotions, save images with bounding boxes.
if len(class_ids) > 0:
found = True
new_filename = 'yolo_' + filename
file_path = os.path.join(here, app.config['UPLOAD_FOLDER'],
new_filename)
yolo_save_img(image,
class_ids,
boxes,
labels,
confidences,
COLORS,
file_path=file_path)
# help function to format result sentences.
def and_syntax(alist):
if len(alist) == 1:
alist = "".join(alist)
return alist
elif len(alist) == 2:
alist = " and ".join(alist)
return alist
elif len(alist) > 2:
alist[-1] = "and " + alist[-1]
alist = ", ".join(alist)
return alist
else:
return
# confidences: rounding and changing to percent, putting in function
format_confidences = []
for percent in confidences:
format_confidences.append(str(round(percent * 100)) + '%')
format_confidences = and_syntax(format_confidences)
# labels: sorting and capitalizing, putting into function
labels = set(labels)
labels = [oreo.capitalize() for oreo in labels]
labels = and_syntax(labels)
# return template with data
return render_template('results.html',
confidences=format_confidences,
labels=labels,
old_filename=filename,
filename=new_filename)
else:
found = False
return render_template('results.html',
labels='no Oreo',
old_filename=filename,
filename=filename)
def get_prediction(image_folder_root,
LABELS,
cfg_path,
weight_path,
confidence_level=0.5):
"""
Run every image in a given folder through darknet.
Return a list of dictionaries with pertinent information
about each image: image_path, class_ids, labels, boxes,
true_labels, and true_boxes.
"""
# Loads YOLO into Python
net = get_yolo_net(cfg_path, weight_path)
# np.random.seed(42)
# colors = np.random.randint(0, 255, size=(len(LABELS), 3), dtype='uint8')
# Lists all files in the directory and splits them into two lists:
# one for images and one for txt files
files = os.listdir(image_folder_root)
image_paths = sorted(
[os.path.join(image_folder_root, f) for f in files if '.jpg' in f])
txt_paths = sorted(
[os.path.join(image_folder_root, f) for f in files if '.txt' in f])
# Loops over each image and txt file in the directory
results = []
for image_path, txt_path in zip(image_paths, txt_paths):
try:
# Get image height and width
image = cv2.imread(image_path)
(H, W) = image.shape[:2]
# Get darknet prediction data
class_ids, labels, boxes, confidences = yolo_forward(
net, LABELS, image, confidence_level)
print(
f"SUCCESS: Predicted Class IDs: {class_ids}; Labels: {labels}; for image: {image_path}.\n"
)
except Exception: # Catch occasional errors
print(f"ERROR: This image had an error: {image_path}")
continue
# Reads ground truth data from txt file
with open(txt_path, "r") as f:
txt_labels = f.readlines()
# Splits data into two lists: labels and boxes
true_labels = [int(label.split()[0]) for label in txt_labels]
true_boxes = [label.split()[1:] for label in txt_labels]
# Convert boxes from YOLO dimensions to standard dimensions
for i, box in enumerate(true_boxes):
box = [float(num) for num in box]
true_boxes[i] = yolo_to_standard_dims(box, H, W)
# Adds pertinent information to a dictionary and adds the dictionary to the return list
result = {
'image_path': image_path,
'class_ids': class_ids,
'labels': labels,
'boxes': boxes,
'confidences': confidences,
'true_labels': true_labels,
'true_boxes': true_boxes
}
results.append(result)
return results
def evaluate():
labels = ["Mom's Car"]
positive_img_dir = os.path.join(os.getcwd(), 'labeled_car_data')
negative_img_dir = os.path.join(os.getcwd(), 'unlabeled_car_data')
confidence_level = 0.5
threshold = 0.3
img_list = []
total_counter = 0
counter = 0
y_actu = []
y_pred = []
net_clement = get_yolo_net('yolov3.cfg', 'yolov3_final.weights')
for img in os.listdir(positive_img_dir):
if img.endswith('.JPG'):
image = cv2.imread(os.path.join(positive_img_dir, img))
img_detect = yolo_forward(net_clement, labels, image,
confidence_level, threshold)
total_counter += 1
if img_detect == ([], [], [], []):
y_actu.append(1)
y_pred.append(0)
continue
else:
img_list.append(img_detect)
print(img_detect)
counter += 1
y_actu.append(1)
y_pred.append(1)
else:
continue
for img in os.listdir(negative_img_dir):
if img.endswith('.JPG'):
image = cv2.imread(os.path.join(negative_img_dir, img))
img_detect = yolo_forward(net_clement, labels, image,
confidence_level, threshold)
total_counter += 1
if img_detect == ([], [], [], []):
y_actu.append(0)
y_pred.append(0)
continue
else:
img_list.append(img_detect)
print(img_detect)
counter += 1
y_actu.append(0)
y_pred.append(1)
else:
continue
print('\n{} images detected out of {} total'.format(
counter, total_counter))
moms_car_counter = 0
for i in img_list:
moms_car_counter += len(i[0])
print("Mom's car detected {} times".format(moms_car_counter))
return y_actu, y_pred