update

ODRS/data_utils/augmentation/aug_bbox.py

@@ -0,0 +1,160 @@
+import cv2
+import numpy as np
+import random
+from aug_image import noisy_salt_and_pepper_img, noisy_gauss_img, brightness_img
+from aug_image import blur_img, saturation_img
+from aug_image import flip_horizontol_img, flip_vertical_img, rotate_img
+from PIL import Image
+
+def get_boxes(img, annotation_path):
+    with open(annotation_path, 'r') as annotation_file:
+        lines = annotation_file.readlines()
+
+    boxes = []
+    for line in lines:
+        data = line.strip().split()
+        class_id, x_center, y_center, width, height = map(float, data)
+        x_center *= img.shape[1]
+        y_center *= img.shape[0]
+        width *= img.shape[1]
+        height *= img.shape[0]
+        x1 = int(x_center - width / 2)
+        y1 = int(y_center - height / 2)
+        x2 = int(x_center + width / 2)
+        y2 = int(y_center + height / 2)
+        boxes.append((x1, y1, x2, y2))
+    return boxes
+
+
+def flip_horizontol_within_box(image_path, annotation_path):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        box_region = Image.fromarray(cv2.cvtColor(box_region, cv2.COLOR_BGR2RGB))
+        flip_box_h = flip_horizontol_img(box_region)
+        flip_box_h = cv2.cvtColor(np.array(flip_box_h), cv2.COLOR_RGB2BGR)
+        image[y1:y2, x1:x2] = flip_box_h
+    cv2.imwrite('image.jpg', image)
+
+
+def flip_vertical_within_box(image_path, annotation_path):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        box_region = Image.fromarray(cv2.cvtColor(box_region, cv2.COLOR_BGR2RGB))
+        flip_box_w = flip_vertical_img(box_region)
+        flip_box_w = cv2.cvtColor(np.array(flip_box_w), cv2.COLOR_RGB2BGR)
+        image[y1:y2, x1:x2] = flip_box_w
+    cv2.imwrite('image.jpg', image)
+
+
+def rotate_within_box(image_path, annotation_path, value):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        box_region = Image.fromarray(cv2.cvtColor(box_region, cv2.COLOR_BGR2RGB))
+        rotate_box = rotate_img(box_region, value)
+        rotate_box = cv2.cvtColor(np.array(rotate_box), cv2.COLOR_RGB2BGR)
+        image[y1:y2, x1:x2] = rotate_box
+    cv2.imwrite('image.jpg', image)
+
+
+def noisy_salt_and_pepper_within_box(image_path, annotation_path,  noise_level):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        noisy_box = noisy_salt_and_pepper_img(box_region, noise_level)
+        image[y1:y2, x1:x2] = noisy_box
+    cv2.imwrite('image.jpg', image)
+
+
+def noisy_gauss_within_box(image_path, annotation_path,  std):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        noisy_box = noisy_gauss_img(box_region, std)
+        image[y1:y2, x1:x2] = noisy_box
+    # cv2.imwrite('image.jpg', image)
+
+
+def brightness_within_box(image_path, annotation_path, std):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        image_box_region = Image.fromarray(cv2.cvtColor(box_region, cv2.COLOR_BGR2RGB))
+        brightness_box = brightness_img(image_box_region, std)
+        brightness_box = cv2.cvtColor(np.array(brightness_box), cv2.COLOR_RGB2BGR)
+        image[y1:y2, x1:x2] = brightness_box
+    # cv2.imwrite('image.jpg', image)
+
+
+def blur_within_box(image_path, annotation_path, blur_radius):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        image_box_region = Image.fromarray(cv2.cvtColor(box_region, cv2.COLOR_BGR2RGB))
+        brightness_box = blur_img(image_box_region, blur_radius)
+        brightness_box = cv2.cvtColor(np.array(brightness_box), cv2.COLOR_RGB2BGR)
+        image[y1:y2, x1:x2] = brightness_box
+    # cv2.imwrite('image.jpg', image)
+
+
+def saturation_within_box(image_path, annotation_path, blur_radius):
+    image = cv2.imread(image_path)
+    boxes = get_boxes(image, annotation_path)
+    for box in boxes:
+        x1, y1, x2, y2 = box
+        box_width = x2 - x1
+        box_height = y2 - y1
+        box_region = image[y1:y2, x1:x2]
+        image_box_region = Image.fromarray(cv2.cvtColor(box_region, cv2.COLOR_BGR2RGB))
+        saturation_box = saturation_img(image_box_region, blur_radius)
+        saturation_box = cv2.cvtColor(np.array(saturation_box), cv2.COLOR_RGB2BGR)
+        image[y1:y2, x1:x2] = saturation_box
+    # cv2.imwrite('image.jpg', image)
+
+
+
+
+
+
+
+# image_path = '/media/space/ssd_1_tb_evo_sumsung/Work/Warp-D/test/images/Monitoring_photo_2_test_25-Mar_11-09-46.jpg'
+# annotation_path = '/media/space/ssd_1_tb_evo_sumsung/Work/Warp-D/test/labels/Monitoring_photo_2_test_25-Mar_11-09-46.txt'
+# # noisy_gauss_within_box(image_path, annotation_path, 17)
+# # noisy_salt_and_pepper_within_box(image_path, annotation_path, 0.7)
+# # brightness_within_box(image_path, annotation_path, 0.7)
+# # blur_within_box(image_path, annotation_path, 1.5)
+# # saturation_within_box(image_path, annotation_path, 3)
+
+# # flip_horizontol_within_box(image_path, annotation_path)
+# # flip_vertical_within_box(image_path, annotation_path)
+# rotate_within_box(image_path, annotation_path, 75)

ODRS/data_utils/augmentation/aug_image.py

@@ -0,0 +1,96 @@
+from PIL import Image, ImageEnhance, ImageFilter
+import numpy as np
+
+
+def flip_horizontol_img(img):
+    img_horizontal_flip = img.transpose(Image.FLIP_LEFT_RIGHT)
+    return img_horizontal_flip
+
+
+def flip_vertical_img(img):
+    img_vertical_flip = img.transpose(Image.FLIP_TOP_BOTTOM)
+    return img_vertical_flip
+
+
+def rotate_img(img, value, expand=False):
+    img_rotated = img.rotate(value, expand=expand)
+    return img_rotated
+
+
+def brightness_img(img, brightness_factor):
+    enhancer = ImageEnhance.Brightness(img)
+    img_brightened = enhancer.enhance(brightness_factor)
+    return img_brightened
+
+
+def noisy_gauss_img(img, std):
+    np_img = np.array(img)
+    gauss = np.random.normal(0, std, np_img.shape)
+    noisy_gauss_img = np_img + gauss
+    noisy_gauss_img = np.clip(noisy_gauss_img, 0, 255).astype(np.uint8)
+    noisy_gauss_img = Image.fromarray(noisy_gauss_img)
+    return noisy_gauss_img
+
+
+def noisy_salt_and_pepper_img(img, noise_level):
+    np_img = np.array(img)
+    amount = int(np.ceil(noise_level * np_img.size / 3))
+    coords = [np.random.randint(0, i - 1, amount) for i in np_img.shape]
+    np_img[coords[0], coords[1]] = 255
+    coords = [np.random.randint(0, i - 1, amount) for i in np_img.shape]
+    np_img[coords[0], coords[1]] = 0
+    noisy_salt_pepper_img = Image.fromarray(np_img)
+    return noisy_salt_pepper_img
+
+
+def blur_img(img, blur_radius):
+    return img.filter(ImageFilter.GaussianBlur(radius=blur_radius))
+
+
+def saturation_img(img, saturation_factor):
+    enhancer = ImageEnhance.Color(img)
+    image_saturation = enhancer.enhance(saturation_factor)
+    return image_saturation
+
+
+
+
+img = Image.open('/media/space/ssd_1_tb_evo_sumsung/exp_pad/dataset/train_dataset/moja/train/images/16-Aug_12-29-54.png')
+
+
+
+# Adjust the saturation level (1.0 is the original saturation)
+# saturation_factor = 10  # You can change this value to adjust saturation (e.g., 0.5 for less saturation)
+# image_saturation = saturation_img(img, saturation_factor)
+# image_saturation.save('modified_image.jpg')
+
+# blur_radius = 1.5  # You can change this value to adjust the blur level
+# blurred_image = blur_img(img, blur_radius)
+
+# # Save the blurred image
+# blurred_image.save('blurred_image.png')
+
+
+# noisy_salt_pepper_img = noisy_salt_and_pepper(img, 0.2)
+# noisy_salt_pepper_img.save('noisy_salt_pepper_img.png')
+
+
+# img_noisy_gauss = noisy_gauss_img(img, 40)
+# img_noisy_gauss.save('noisy_gauss.png')
+
+
+# img_brightness = brightness_img(img, 1.5)
+# img_brightness.save('brightness.png')
+
+# img_horizontal_flip = flip_horizontol(img)
+# img_horizontal_flip.save('horizontal_flip.png')
+
+# img_vertical_flip = flip_vertical(img)
+# img_vertical_flip.save('img_vertical_flip.png')
+
+# # img_diagonal_flip_main = img.transpose(Image.Transpose.ROTATE_180)  # Поворот на 90 градусов
+# # img_diagonal_flip_main.save('diagonal_flip_main.png')
+
+
+# img_rotated = rotate(img, 75, expand=False)
+# img_rotated.save('rotated_image.png')

ODRS/data_utils/prepare_ssd.py

@@ -92,54 +92,4 @@ def create_ssd_json(path_folder, txt_path):
                 except Exception as e:
                     print(e)
 
-        save_as_json(Path(os.path.dirname(path_folder)) / f'{path_folder.name}.json', dataset)
-
-
-
-def resize_images_and_annotations(data_path, img_size):
-    size = img_size if img_size <= 300 else 300
-    path = Path(data_path)
-    folder_names = [folder.name for folder in path.iterdir() if folder.is_dir()]
-    for name in folder_names:
-        folder_path = path / name
-        images_path = os.path.join(folder_path, 'images')
-        labels_path = os.path.join(folder_path, 'labels')
-
-        for image_name in tqdm(os.listdir(images_path), desc=f'Resize {name} images'):
-            image_path = os.path.join(images_path, image_name)
-            label_path = os.path.join(labels_path, image_name.replace('.jpg', '.txt'))
-
-            with Image.open(image_path) as img:
-                original_width, original_height = img.size
-
-                if original_width > size or original_height > size:
-                    img = img.resize((size, size))
-
-                    if os.path.exists(label_path):
-                        with open(label_path, 'r') as file:
-                            lines = file.readlines()
-
-                        with open(label_path, 'w') as file:
-                            for line in lines:
-                                parts = line.split()
-                                if len(parts) == 5:
-                                    x_center = float(parts[1]) * original_width
-                                    y_center = float(parts[2]) * original_height
-                                    width = float(parts[3]) * original_width
-                                    height = float(parts[4]) * original_height
-
-                                    x_center *= size / original_width
-                                    y_center *= size / original_height
-                                    width *= size / original_width
-                                    height *= size / original_height
-
-                                    x_center /= size
-                                    y_center /= size
-                                    width /= size
-                                    height /= size
-
-                                    file.write(f"{parts[0]} {x_center} {y_center} {width} {height}\n")
-
-                    img.save(image_path)
-
-# resize_images_and_annotations('/media/space/ssd_1_tb_evo_sumsung/ITMO/ODRS/user_datasets/Warp-D_voc/test')
+        save_as_json(Path(os.path.dirname(path_folder)) / f'{path_folder.name}.json', dataset)

ODRS/data_utils/resize_image.py

@@ -0,0 +1,60 @@
+import os
+from tqdm import tqdm
+from pathlib import Path
+from PIL import Image
+
+def resize_images_and_annotations(data_path, img_size):
+    if isinstance(img_size, int):
+        width = height = img_size
+    elif isinstance(img_size, tuple) and len(img_size) == 2:
+        width, height = img_size
+    else:
+        raise ValueError("Invalid img_size format. Please provide either an integer or a tuple of two integers.")
+
+    path = Path(data_path)
+    folder_names = [folder.name for folder in path.iterdir() if folder.is_dir()]
+
+    for name in folder_names:
+        folder_path = path / name
+        images_path = os.path.join(folder_path, 'images')
+        labels_path = os.path.join(folder_path, 'labels')
+
+        for image_name in tqdm(os.listdir(images_path), desc=f'Resize {name} images'):
+            image_path = os.path.join(images_path, image_name)
+            label_path = os.path.join(labels_path, image_name.replace('.jpg', '.txt'))
+
+            with Image.open(image_path) as img:
+                original_width, original_height = img.size
+
+                if original_width != width or original_height != height:
+                    img = img.resize((width, height))
+
+                    if os.path.exists(label_path):
+                        with open(label_path, 'r') as file:
+                            lines = file.readlines()
+
+                        with open(label_path, 'w') as file:
+                            for line in lines:
+                                parts = line.split()
+                                if len(parts) == 5:
+                                    x_center = float(parts[1]) * original_width
+                                    y_center = float(parts[2]) * original_height
+                                    box_width = float(parts[3]) * original_width
+                                    box_height = float(parts[4]) * original_height
+
+                                    x_center *= width / original_width
+                                    y_center *= height / original_height
+                                    box_width *= width / original_width
+                                    box_height *= height / original_height
+
+                                    x_center /= width
+                                    y_center /= height
+                                    box_width /= width
+                                    box_height /= height
+
+                                    file.write(f"{parts[0]} {x_center} {y_center} {box_width} {box_height}\n")
+
+                    img.save(image_path)
+
+
+# resize_images_and_annotations('/media/space/ssd_1_tb_evo_sumsung/Work/Warp-D', (640, 480))

ODRS/train_utils/config/custom_config.yaml

@@ -1,10 +1,10 @@
-BATCH_SIZE: 10
+BATCH_SIZE: 20
 CLASSES: classes.txt
-DATA_PATH: Warp-D
+DATA_PATH: /media/space/ssd_1_tb_evo_sumsung/Work/Warp-D
 EPOCHS: 2
 GPU_COUNT: 1
-IMG_SIZE: 200
-MODEL: yolov5l
+IMG_SIZE: 300
+MODEL: ssd
 SELECT_GPU: 0
 
 CONFIG_PATH: dataset.yaml