Upload folder using huggingface_hub

README.md

@@ -1,3 +1,73 @@
----
-license: mit
----
+---
+license: mit
+task_categories:
+  - object-detection
+tags:
+  - traffic-signs
+  - vietnam
+  - yolo
+  - residential-zone
+  - R420
+  - R421
+size_categories:
+  - n<1K
+---
+
+# VietSpeedYOLO — Residential zone traffic signs (R420/R421)
+
+YOLO-format object detection dataset for **Vietnam residential-zone traffic signs**: **R420** (Bắt đầu khu dân cư) and **R421** (Hết khu dân cư). Use for speed-zone detection, ADAS, or traffic sign recognition in Vietnam.
+
+## Classes
+
+| ID | Class name     | Sign | Description              |
+|----|----------------|------|--------------------------|
+| 0  | khu_dan_cu     | R420 | Start of residential area |
+| 1  | ngoai_dan_cu   | R421 | End of residential area  |
+
+## Dataset structure
+
+Standard YOLO layout:
+
+- **images/train/** — training images
+- **images/val/** — validation images  
+- **labels/train/** — one `.txt` per image: `class_id x_center y_center width height` (normalized 0–1)
+- **labels/val/** — validation labels
+- **data.yaml** — dataset config (paths + class names)
+
+## Usage
+
+### With Ultralytics YOLO
+
+```bash
+# Clone or download this repo, then:
+yolo detect train data=data.yaml model=yolov8n.pt epochs=80 batch=16 imgsz=640
+```
+
+### With Python
+
+```python
+from ultralytics import YOLO
+model = YOLO("yolov8n.pt")
+model.train(data="data.yaml", epochs=80, batch=16, imgsz=640)
+```
+
+### Loading from Hugging Face
+
+After cloning or downloading the dataset from the Hub, point your training script to the local `data.yaml`. Paths in `data.yaml` are relative to the dataset root (this repo).
+
+## Statistics (example)
+
+- Train: 94 original images (+ optional augmentation for training)
+- Val: 23 images
+- Labels: one text file per image, YOLO format
+
+## License
+
+MIT.
+
+## Citation
+
+If you use this dataset, please cite:
+
+- **Dataset:** [NghiMe/vietspeedyolo](https://huggingface.co/datasets/NghiMe/vietspeedyolo) on Hugging Face
+- **Code:** [nghimestudio/vietspeedyolo](https://github.com/nghimestudio/vietspeedyolo) on GitHub

README_TRAIN.md

@@ -0,0 +1,45 @@
+# Train model nhận diện box + phân loại khu dân cư / ngoài khu dân cư
+
+## Dataset
+- **traffic_residence_2class/** đã có 2 class:
+  - **0: khu_dan_cu** (Bắt đầu khu dân cư - R420)
+  - **1: ngoai_dan_cu** (Hết khu dân cư - R421)
+- Mỗi file label `.txt`: `class_id x_center y_center width height` (chuẩn hóa 0–1).
+- Một model YOLO vừa **tìm tọa độ box** vừa **phân loại** 0 hay 1.
+
+## Train
+Từ thư mục `archive/`:
+
+```bash
+# Cách 1: script
+python3 train_residence_2class.py
+
+# Cách 2: lệnh yolo
+yolo detect train data=traffic_residence_2class/data.yaml model=yolov8n.pt epochs=80 batch=16 imgsz=640
+```
+
+Sau khi train xong, weights nằm ở: **runs/detect/residence_2class/weights/best.pt** (hoặc runs/detect/train/ nếu dùng lệnh yolo trực tiếp).
+
+## Dùng model (inference)
+Ảnh mới → model trả về **tọa độ từng box** và **class (0 hoặc 1)**:
+
+```bash
+yolo detect predict model=runs/detect/residence_2class/weights/best.pt source=path/to/anh.jpg save
+```
+
+Hoặc Python:
+
+```python
+from ultralytics import YOLO
+model = YOLO("runs/detect/residence_2class/weights/best.pt")
+results = model.predict("anh.jpg", conf=0.25)
+for r in results:
+    boxes = r.boxes.xyxy.cpu().numpy()   # [x1,y1,x2,y2] từng box
+    classes = r.boxes.cls.cpu().numpy()  # 0 hoặc 1
+    for box, cls in zip(boxes, classes):
+        x1, y1, x2, y2 = box
+        label = "khu_dan_cu" if cls == 0 else "ngoai_dan_cu"
+        # dùng (x1,y1,x2,y2) và label
+```
+
+Kết quả: mỗi detection có **vị trí box** (x1,y1,x2,y2) và **nhãn** 0 (khu dân cư) hay 1 (ngoài khu dân cư).

data.yaml

@@ -0,0 +1,7 @@
+path: /Users/thanhnguyen/Downloads/archive/traffic_residence_2class
+train: images/train
+val: images/val
+
+names:
+  0: khu_dan_cu
+  1: ngoai_dan_cu