fix augmentation - correct bboxes to be within borders+margin

Add classes = 80 as default number in yaml file

README.md

@@ -36,13 +36,12 @@ Linux
 ```
 
 **3. Fix possible problems**
-* Windows:
   * cv2.error: OpenCV(4.10.0) ...\window.cpp:1301: error: (-2:Unspecified error)
       ```
       pip uninstall opencv-python 
       pip install opencv-python
       ```
-  * fbgemm.dll  error
+  * fbgemm.dll  error (Windows specific)
       ```
       copypaste libomp140.x86_64.dll to C:\Windows\System32
       ```

classes.json

@@ -1,61 +1,61 @@
  [
     {
       "Id": 0,
-      "Name": "Броньована техніка",
+      "Name": "Armored-Vehicle",
       "Color": "#80FF0000",
       "ColorBrush": "#80FF0000"
     },
     {
       "Id": 1,
-      "Name": "Вантажівка",
+      "Name": "Truck",
       "Color": "#8000FF00",
       "ColorBrush": "#8000FF00"
     },
     {
       "Id": 2,
-      "Name": "Машина легкова",
+      "Name": "Vehicle",
       "Color": "#800000FF",
       "ColorBrush": "#800000FF"
     },
     {
       "Id": 3,
-      "Name": "Артилерія",
+      "Name": "Artillery",
       "Color": "#80FFFF00",
       "ColorBrush": "#80FFFF00"
     },
     {
       "Id": 4,
-      "Name": "Тінь від техніки",
+      "Name": "Shadow",
       "Color": "#80FF00FF",
       "ColorBrush": "#80FF00FF"
     },
     {
       "Id": 5,
-      "Name": "Окопи",
+      "Name": "Trenches",
       "Color": "#8000FFFF",
       "ColorBrush": "#8000FFFF"
     },
     {
       "Id": 6,
-      "Name": "Військовий",
+      "Name": "Military-men",
       "Color": "#80000000",
       "ColorBrush": "#80000000"
     },
     {
       "Id": 7,
-      "Name": "Накати",
+      "Name": "Tyre-tracks",
       "Color": "#80800000",
       "ColorBrush": "#80800000"
     },
     {
       "Id": 8,
-      "Name": "Танк з додатковим захистом",
+      "Name": "Additional-armored-tank",
       "Color": "#80008000",
       "ColorBrush": "#80008000"
     },
     {
       "Id": 9,
-      "Name": "Дим",
+      "Name": "Smoke",
       "Color": "#80000080",
       "ColorBrush": "#80000080"
     }

convert-annotations.py

@@ -4,8 +4,6 @@ import xml.etree.cElementTree as et
 from pathlib import Path
 import cv2
 
-labels_dir = 'labels'
-images_dir = 'images'
 
 tag_size = 'size'
 tag_object = 'object'
@@ -17,9 +15,11 @@ default_class = 1
 image_extensions = ['jpg', 'png', 'jpeg']
 
 
-def convert(folder, read_annotations, ann_format):
+def convert(folder, dest_folder, read_annotations, ann_format):
-    os.makedirs(images_dir, exist_ok=True)
+    dest_images_dir = os.path.join(dest_folder, 'images')
-    os.makedirs(labels_dir, exist_ok=True)
+    dest_labels_dir = os.path.join(dest_folder, 'labels')
+    os.makedirs(dest_images_dir, exist_ok=True)
+    os.makedirs(dest_labels_dir, exist_ok=True)
 
     for f in os.listdir(folder):
         if not f[-3:] in image_extensions:
@@ -39,8 +39,8 @@ def convert(folder, read_annotations, ann_format):
         except Exception as e:
             print(f'Error conversion for {f}. Error: {e}')
 
-        shutil.copy(os.path.join(folder, f), os.path.join(images_dir, f))
+        shutil.copy(os.path.join(folder, f), os.path.join(dest_images_dir, f))
-        with open(os.path.join(labels_dir, f'{Path(label).stem}.txt'), 'w') as new_label_file:
+        with open(os.path.join(dest_labels_dir, f'{Path(label).stem}.txt'), 'w') as new_label_file:
             new_label_file.writelines(lines)
             new_label_file.close()
         print(f'Image {f} has been processed successfully')
@@ -115,5 +115,5 @@ def rename_images(folder):
 
 
 if __name__ == '__main__':
-    convert('datasets/others/UAVHeightImages', read_bbox_oriented, 'txt')
+    convert('/azaion/datasets/others/cars_height', '/azaion/datasets/converted', read_bbox_oriented, 'txt')
-    convert('datasets/others/UAVimages', read_pascal_voc, 'xml')
+    convert('/azaion/datasets/others/cars', '/azaion/datasets/converted', read_pascal_voc, 'xml')

convert-rknn/install_converter_yolov8.sh

@@ -0,0 +1,41 @@
+# prerequisites 
+sudo apt install python3.12-venv python-is-python3 
+if [ ! -f azaion.pt ]; then
+    echo "----ERROR!----"
+    echo "Script requires azaion.pt file model in this folder!!!"
+    echo "----ERROR!----"
+    exit
+fi
+
+# convert PT to ONNX
+git clone https://github.com/airockchip/ultralytics_yolov8
+cd ultralytics_yolov8
+python -m venv env
+source env/bin/activate
+pip install .
+
+cp ../azaion.pt .
+sed -i -E "s/(model:).*$/\1 azaion.pt/" ultralytics/cfg/default.yaml
+export PYTHONPATH=./
+python ./ultralytics/engine/exporter.py
+cp azaion.onnx ../
+cd ..
+
+exit 
+
+# convert ONNX to RKNN
+wget -c https://mirrors.bfsu.edu.cn/anaconda/miniconda/Miniconda3-latest-Linux-x86_64.sh
+chmod +x Miniconda3-latest-Linux-x86_64.sh
+bash Miniconda3-latest-Linux-x86_64.sh
+source ~/miniconda3/bin/activate
+conda create -n toolkit2 python=3.11
+conda activate toolkit2
+git clone https://github.com/rockchip-linux/rknn-toolkit2.git
+cd rknn-toolkit2/rknn-toolkit2/packages
+pip install -r requirements_cp311-1.6.0.txt
+pip install rknn_toolkit2-1.6.0+81f21f4d-cp311-cp311-linux_x86_64.whl
+cd ~/opi5-rknn
+git clone https://github.com/airockchip/rknn_model_zoo.git
+cd rknn_model_zoo/examples/yolov8/python
+python convert.py ~/azaion/models/azaion-2024-08-13.onnx rk3588 i8
+cp ../model/yolov8.rknn ~/azaion/models/azaion-2024-08-13.rknn # Output file form convert script is hard-coded...

preprocessing.py

@@ -11,10 +11,35 @@ from constants import (data_images_dir, data_labels_dir, processed_images_dir, p
 from dto.imageLabel import ImageLabel
 
 
+def correct_bboxes(labels):
+    margin = 0.0005
+    min_size = 0.01
+    res = []
+    for bboxes in labels:
+        x = bboxes[0]
+        y = bboxes[1]
+        half_width = 0.5*bboxes[2]
+        half_height = 0.5*bboxes[3]
+
+        # calc how much bboxes are outside borders ( +small margin ).
+        # value should be negative. If it's positive, then put 0, as no correction
+        w_diff = min( (1 - margin) - (x + half_width), (x - half_width) - margin, 0 )
+        w = bboxes[2] + 2*w_diff
+        if w < min_size:
+            continue
+        h_diff = min( (1 - margin) - (y + half_height), ((y - half_height) - margin), 0)
+        h = bboxes[3] + 2 * h_diff
+        if h < min_size:
+            continue
+        res.append([x, y, w, h, bboxes[4]])
+    return res
+    pass
+
+
 def image_processing(img_ann: ImageLabel) -> [ImageLabel]:
     transforms = [
         A.Compose([A.HorizontalFlip(always_apply=True)],
-                  bbox_params=A.BboxParams(format='yolo')),
+                  bbox_params=A.BboxParams(format='yolo', )),
         A.Compose([A.RandomBrightnessContrast(always_apply=True)],
                   bbox_params=A.BboxParams(format='yolo')),
         A.Compose([A.SafeRotate(limit=90, always_apply=True)],
@@ -33,9 +58,12 @@ def image_processing(img_ann: ImageLabel) -> [ImageLabel]:
     ]
 
     results = []
+    labels = correct_bboxes(img_ann.labels)
+    if len(labels) == 0 and len(img_ann.labels) != 0:
+        print('no labels but was!!!')
     for i, transform in enumerate(transforms):
         try:
-            res = transform(image=img_ann.image, bboxes=img_ann.labels)
+            res = transform(image=img_ann.image, bboxes=labels)
             path = Path(img_ann.image_path)
             name = f'{path.stem}_{i + 1}'
             img = ImageLabel(
@@ -51,9 +79,6 @@ def image_processing(img_ann: ImageLabel) -> [ImageLabel]:
 
 
 def write_result(img_ann: ImageLabel):
-    os.makedirs(os.path.dirname(img_ann.image_path), exist_ok=True)
-    os.makedirs(os.path.dirname(img_ann.labels_path), exist_ok=True)
-
     cv2.imencode('.jpg', img_ann.image)[1].tofile(img_ann.image_path)
     print(f'{img_ann.image_path} written')
 
@@ -92,6 +117,8 @@ def process_image(img_ann):
 
 
 def main():
+    os.makedirs(processed_images_dir, exist_ok=True)
+    os.makedirs(processed_labels_dir, exist_ok=True)
     while True:
         processed_images = set(f.name for f in os.scandir(processed_images_dir))
         images = []

scripts/init-sftp.sh

@@ -3,18 +3,38 @@ apt update
 apt upgrade
 apt install -y ssh
 
+addgroup sftp
 adduser azaionsftp
+usermod -G sftp azaionsftp
+usermod -a -G sftp zxsanny
 
-mkdir /azaion-media/sftphome
+chown root:root /home/azaionsftp
-chown -R azaionsftp:azaionsftp /azaion-media/sftphome/
+
-chmod -R 755 /azaion-media/sftphome/
+cd /home/azaionsftp
+mkdir datasets
+chown -R azaionsftp:azaionsftp datasets
+mount -o bind /azaion/datasets datasets
+
+chown -R zxsanny:sftp /azaion/data
+mkdir data
+chown -R azaionsftp:azaionsftp data
+mount -o bind /azaion/data data
+
+chown -R zxsanny:sftp /azaion/data-processed
+mkdir data-processed
+chown -R azaionsftp:azaionsftp data-processed
+mount -o bind /azaion/data-processed data-processed
+
+
+chmod -R 755 /home/azaionsftp/
 
 cat <<EOT >> /etc/ssh/sshd_config
-Match Group azaionsftp
+Match Group sftp
     ChrootDirectory %h
+    PasswordAuthentication yes
+    AllowTcpForwarding no
     X11Forwarding no
     ForceCommand internal-sftp
-    AllowTcpForwarding no
-    PasswordAuthentication yes
 EOT
 service ssh restart
+

train.py

@@ -18,6 +18,8 @@ train_set = 70
 valid_set = 20
 test_set = 10
 
+DEFAULT_CLASS_NUM = 80
+
 
 def form_dataset(from_date: datetime):
     makedirs(today_dataset, exist_ok=True)
@@ -89,7 +91,10 @@ def create_yaml():
     lines = ['names:']
     for c in annotation_classes:
         lines.append(f'- {annotation_classes[c].name}')
-    lines.append(f'nc: {len(annotation_classes)}')
+    classes_count = len(annotation_classes)
+    for c in range(DEFAULT_CLASS_NUM - classes_count):
+        lines.append(f'- Class-{c + classes_count + 1}')
+    lines.append(f'nc: {DEFAULT_CLASS_NUM}')
 
     lines.append(f'test: test/images')
     lines.append(f'train: train/images')
@@ -132,17 +137,19 @@ def get_latest_model():
 
 if __name__ == '__main__':
     latest_date, latest_model = get_latest_model()
+    # create_yaml()
     # form_dataset(latest_date)
 
+
     model_name = latest_model if latest_model is not None and path.isfile(latest_model) else 'yolov8m.yaml'
     print(f'Initial model: {model_name}')
     model = YOLO(model_name)
     
-    # cur_folder = path.join(datasets_dir, f'{prefix}2024-06-18')
+    cur_folder = path.join(datasets_dir, f'{prefix}2024-09-19')
 
-    cur_folder = today_dataset
+    # cur_folder = today_dataset
     yaml = abspath(path.join(cur_folder, 'data.yaml'))
-    results = model.train(data=yaml, epochs=100, batch=60, imgsz=640)
+    results = model.train(data=yaml, epochs=100, batch=58, imgsz=640)
 
     shutil.copytree(results.save_dir, path.join(models_dir, today_folder))
     shutil.rmtree('runs')

yolov8.yaml

@@ -0,0 +1,46 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
+
+# Parameters
+nc: 80  # number of classes
+scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
+  # [depth, width, max_channels]
+  n: [0.33, 0.25, 1024]  # YOLOv8n summary: 225 layers,  3157200 parameters,  3157184 gradients,   8.9 GFLOPs
+  s: [0.33, 0.50, 1024]  # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients,  28.8 GFLOPs
+  m: [0.67, 0.75, 768]   # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients,  79.3 GFLOPs
+  l: [1.00, 1.00, 512]   # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
+  x: [1.00, 1.25, 512]   # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
+
+# YOLOv8.0n backbone
+backbone:
+  # [from, repeats, module, args]
+  - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
+  - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
+  - [-1, 3, C2f, [128, True]]
+  - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
+  - [-1, 6, C2f, [256, True]]
+  - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
+  - [-1, 6, C2f, [512, True]]
+  - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
+  - [-1, 3, C2f, [1024, True]]
+  - [-1, 1, SPPF, [1024, 5]]  # 9
+
+# YOLOv8.0n head
+head:
+  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
+  - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
+  - [-1, 3, C2f, [512]]  # 12
+
+  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
+  - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
+  - [-1, 3, C2f, [256]]  # 15 (P3/8-small)
+
+  - [-1, 1, Conv, [256, 3, 2]]
+  - [[-1, 12], 1, Concat, [1]]  # cat head P4
+  - [-1, 3, C2f, [512]]  # 18 (P4/16-medium)
+
+  - [-1, 1, Conv, [512, 3, 2]]
+  - [[-1, 9], 1, Concat, [1]]  # cat head P5
+  - [-1, 3, C2f, [1024]]  # 21 (P5/32-large)
+
+  - [[15, 18, 21], 1, Detect, [nc]]  # Detect(P3, P4, P5)

yolov8m.yaml

@@ -1,40 +0,0 @@
-# Ultralytics YOLO 🚀, GPL-3.0 license
-
-# Parameters
-nc: 50  # number of classes
-depth_multiple: 0.67  # scales module repeats
-width_multiple: 0.75  # scales convolution channels
-
-# YOLOv8.0m backbone
-backbone:
-  # [from, repeats, module, args]
-  - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
-  - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
-  - [-1, 3, C2f, [128, True]]
-  - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
-  - [-1, 6, C2f, [256, True]]
-  - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
-  - [-1, 6, C2f, [512, True]]
-  - [-1, 1, Conv, [768, 3, 2]]  # 7-P5/32
-  - [-1, 3, C2f, [768, True]]
-  - [-1, 1, SPPF, [768, 5]]  # 9
-
-# YOLOv8.0m head
-head:
-  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
-  - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
-  - [-1, 3, C2f, [512]]  # 13
-
-  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
-  - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
-  - [-1, 3, C2f, [256]]  # 17 (P3/8-small)
-
-  - [-1, 1, Conv, [256, 3, 2]]
-  - [[-1, 12], 1, Concat, [1]]  # cat head P4
-  - [-1, 3, C2f, [512]]  # 20 (P4/16-medium)
-
-  - [-1, 1, Conv, [512, 3, 2]]
-  - [[-1, 9], 1, Concat, [1]]  # cat head P5
-  - [-1, 3, C2f, [768]]  # 23 (P5/32-large)
-
-  - [[15, 18, 21], 1, Detect, [nc]]  # Detect(P3, P4, P5)