Merge pull request #2 from maycuatroi/develop

YOLOv8 develop

.gitignore

@@ -132,4 +132,5 @@ dmypy.json
 .github/templates/*
 .env
 .idea
-data/
+data/
+weights/

.vscode/settings.json

@@ -1,4 +1,3 @@
-
 {
     "python.envFile": "${workspaceFolder}/.env",
     "python.autoComplete.extraPaths": [
@@ -9,5 +8,7 @@
     ],
     "files.exclude": {
         "**/COCO": true
-    }
+    },
+    "python.terminal.activateEnvironment": true,
+    "python.languageServer": "Pylance"
 }

README.md

@@ -27,6 +27,43 @@ pip install evo-science
     model.calculate_coefficients(x=x)
 ```
 
+## YOLO Object Detection
+
+The library includes a comprehensive implementation of YOLO (You Only Look Once) object detection models, including YOLOv8. The implementation features:
+
+- Full YOLOv8 architecture with backbone, neck (FPN), and detection head
+- Distributed training support
+- Real-time object detection with webcam
+- Model profiling and EMA (Exponential Moving Average) support
+- Custom loss functions including DFL (Distribution Focal Loss)
+
+### YOLO Example
+
+```python
+from evo_science.packages.yolo.yolo_v8 import YoloV8
+
+# Initialize YOLOv8-nano model
+model = YoloV8.yolo_v8_n(num_classes=80)  # 80 classes for COCO dataset
+
+# For training
+from evo_science.packages.yolo.modules.trainer import Trainer, TrainerConfig
+
+config = TrainerConfig(
+    data_dir="path/to/coco",
+    batch_size=32,
+    epochs=300,
+    input_size=640
+)
+
+trainer = Trainer(model, config)
+trainer.train()
+
+# For real-time detection using webcam
+from evo_science.packages.yolo.modules.demo import demo
+
+demo(input_size=640, model=model)
+```
+
 ## Development
 
 Read the [CONTRIBUTING.md](CONTRIBUTING.md) file.

evo_science/datasets/abstract_dataset.py

@@ -0,0 +1,5 @@
+from torch.utils import data
+
+
+class AbstractDataset(data.Dataset):
+    pass

evo_science/datasets/augmenters/__init__.py

@@ -0,0 +1,3 @@
+from evo_science.datasets.augmenters.image_augmenter import ImageAugmenter
+
+__all__ = ["ImageAugmenter"]

evo_science/datasets/augmenters/image_augmenter.py

@@ -0,0 +1,54 @@
+from typing import Tuple
+
+import cv2
+import numpy as np
+
+
+class ImageAugmenter:
+    def __init__(self):
+        self.transform = None
+        try:
+            import albumentations as A
+
+            transforms = [
+                A.Blur(p=0.01),
+                A.CLAHE(p=0.01),
+                A.ToGray(p=0.01),
+                A.MedianBlur(p=0.01),
+            ]
+            self.transform = A.Compose(transforms, A.BboxParams("yolo", ["class_labels"]))
+        except ImportError:
+            pass
+
+    def __call__(
+        self, image: np.ndarray, box: np.ndarray, cls: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
+        if self.transform:
+            x = self.transform(image=image, bboxes=box, class_labels=cls)
+            image = x["image"]
+            box = np.array(x["bboxes"])
+            cls = np.array(x["class_labels"])
+        return image, box, cls
+
+    @staticmethod
+    def augment_hsv(image: np.ndarray) -> None:
+        h, s, v = 0.015, 0.7, 0.4
+        r = np.random.uniform(-1, 1, 3) * [h, s, v] + 1
+        hue, sat, val = cv2.split(cv2.cvtColor(image, cv2.COLOR_BGR2HSV))
+
+        x = np.arange(0, 256, dtype=r.dtype)
+        lut_h = ((x * r[0]) % 180).astype("uint8")
+        lut_s = np.clip(x * r[1], 0, 255).astype("uint8")
+        lut_v = np.clip(x * r[2], 0, 255).astype("uint8")
+
+        hsv = cv2.merge((cv2.LUT(hue, lut_h), cv2.LUT(sat, lut_s), cv2.LUT(val, lut_v)))
+        cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR, dst=image)
+
+    @staticmethod
+    def mix_up(
+        image1: np.ndarray, box1: np.ndarray, image2: np.ndarray, box2: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray]:
+        alpha = np.random.beta(32.0, 32.0)
+        image = (image1 * alpha + image2 * (1 - alpha)).astype(np.uint8)
+        box = np.concatenate((box1, box2), 0)
+        return image, box