feat: add SmokeGuard models for HuggingFace

README.md

@@ -1,3 +1,61 @@
 ---
 license: mit
+tags:
+- yolov5
+- smoking-detection
+- object-detection
+- onnx
+- pytorch
 ---
+
+# SmokeGuard Models
+
+YOLOv5 models untuk deteksi aktivitas merokok secara real-time.
+
+## Models
+
+| Model | Format | Size | Description |
+|-------|--------|------|-------------|
+| `best_checkpoint.pt` | PyTorch | ~40 MB | Model training terbaik |
+| `last_checkpoint.pt` | PyTorch | ~40 MB | Checkpoint terakhir |
+| `best_checkpoint.onnx` | ONNX (FP32) | ~80 MB | Export ONNX untuk inference |
+| `best_checkpoint_int8.onnx` | ONNX (INT8) | ~20 MB | Quantized untuk CPU inference |
+
+## Usage
+
+### PyTorch
+
+```python
+import torch
+
+model = torch.hub.load("ultralytics/yolov5", "custom", path="best_checkpoint.pt")
+results = model("image.jpg")
+results.show()
+```
+
+### ONNX Runtime
+
+```python
+import onnxruntime as ort
+import numpy as np
+
+session = ort.InferenceSession("best_checkpoint_int8.onnx")
+input_name = session.get_inputs()[0].name
+input_data = np.random.randn(1, 3, 640, 640).astype(np.float32)
+output = session.run(None, {input_name: input_data})
+```
+
+## Quantization
+
+Untuk quantization ONNX model, lihat notebook `quantisize.ipynb`.
+
+## Citation
+
+```bibtex
+@article{IPI4527801,
+  title = "IMPLEMENTASI METODE YOLOv5 PADA SISTEM PENDETEKSI ROKOK DI AREA BEBAS ASAP ROKOK",
+  journal = "Institut Penelitian Matematika, Komputer, Keperawatan, Pendidikan dan Ekonomi (IPM2KPE)",
+  year = "2024",
+  author = "Fathoni, Aliffatul Majid; Zuliarso, Eri"
+}
+```

best_checkpoint.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea3eb0d895c7ad9777f7cb991f4d9f634fd0bea0dc3a672b5a18ed70848b0a56
+size 83890343

best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6f8c39a73f4ff7328482f18a7a48dd1deb2925446d9c92ab3b658ca32f784f6
+size 42234324

best_checkpoint_int8.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b19100820040a6927830526f2f448984d78c518585ae0e773fe8bb6348866b0
+size 21512155

last_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f18948251bdef9e5ff5774d7f8e15fe0d2ac3fd7466a8bdb2b945051bc384467
+size 42234324

quantisize.ipynb

@@ -0,0 +1,172 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "d42d2f51",
+   "metadata": {},
+   "source": [
+    "# ONNX Quantization\n",
+    "\n",
+    "Quantization ONNX model ke INT8 dengan preprocessing."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "411d7be9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import numpy as np\n",
+    "import onnx\n",
+    "from pathlib import Path\n",
+    "from onnxruntime.quantization import quantize_static, QuantType, CalibrationDataReader"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b87dc761",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def get_size_mb(path):\n",
+    "    return os.path.getsize(path) / (1024**2)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a3af9a45",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "ONNX_MODEL = \"/content/drive/MyDrive/models/best_checkpoint.onnx\"\n",
+    "ONNX_PREPROCESSED = str(Path(ONNX_MODEL).with_suffix('')) + \"_preprocessed.onnx\"\n",
+    "ONNX_INT8 = str(Path(ONNX_MODEL).with_suffix('')) + \"_int8.onnx\"\n",
+    "\n",
+    "if not os.path.exists(ONNX_MODEL):\n",
+    "    raise FileNotFoundError(f\"Model not found: {ONNX_MODEL}\")\n",
+    "\n",
+    "print(f\"Input: {Path(ONNX_MODEL).name} ({get_size_mb(ONNX_MODEL):.2f} MB)\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "bde9da41",
+   "metadata": {},
+   "source": [
+    "## 1. Preprocessing"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f89cfbb8",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!python -m onnxruntime.quantization.preprocess --input {ONNX_MODEL} --output {ONNX_PREPROCESSED}\n",
+    "\n",
+    "print(f\"Preprocessed: {get_size_mb(ONNX_PREPROCESSED):.2f} MB\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "433bb1f9",
+   "metadata": {},
+   "source": [
+    "## 2. Calibration Data Reader"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ec3cb386",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "class CalibrationReader(CalibrationDataReader):\n",
+    "    def __init__(self, model_path, num_samples=10):\n",
+    "        model = onnx.load(model_path)\n",
+    "        input_shape = tuple([d.dim_value for d in model.graph.input[0].type.tensor_type.shape.dim])\n",
+    "        self.data = [np.random.randn(*input_shape).astype(np.float32) for _ in range(num_samples)]\n",
+    "        self.input_name = model.graph.input[0].name\n",
+    "        self.enum_index = 0\n",
+    "    \n",
+    "    def get_next(self):\n",
+    "        if self.enum_index >= len(self.data):\n",
+    "            return None\n",
+    "        input_dict = {self.input_name: self.data[self.enum_index]}\n",
+    "        self.enum_index += 1\n",
+    "        return input_dict"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "d1080068",
+   "metadata": {},
+   "source": [
+    "## 3. INT8 Quantization"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "66501947",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "calibration_reader = CalibrationReader(ONNX_PREPROCESSED)\n",
+    "\n",
+    "quantize_static(\n",
+    "    model_input=ONNX_PREPROCESSED,\n",
+    "    model_output=ONNX_INT8,\n",
+    "    calibration_data_reader=calibration_reader,\n",
+    "    weight_type=QuantType.QUInt8,\n",
+    "    optimize_model=False\n",
+    ")\n",
+    "\n",
+    "print(f\"INT8: {get_size_mb(ONNX_INT8):.2f} MB ({(1 - get_size_mb(ONNX_INT8)/get_size_mb(ONNX_MODEL)) * 100:.1f}% reduction)\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "0a20622e",
+   "metadata": {},
+   "source": [
+    "## 4. Summary"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "05d66ea4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(\"=\" * 50)\n",
+    "print(f\"{'Model':<15} {'Size (MB)':<15} {'Reduksi':<10}\")\n",
+    "print(\"-\" * 50)\n",
+    "print(f\"{'Original':<15} {get_size_mb(ONNX_MODEL):.2f}{'':<8} baseline\")\n",
+    "print(f\"{'Preprocessed':<15} {get_size_mb(ONNX_PREPROCESSED):.2f}{'':<8} {(1-get_size_mb(ONNX_PREPROCESSED)/get_size_mb(ONNX_MODEL))*100:.1f}%\")\n",
+    "print(f\"{'INT8':<15} {get_size_mb(ONNX_INT8):.2f}{'':<8} {(1-get_size_mb(ONNX_INT8)/get_size_mb(ONNX_MODEL))*100:.1f}%\")\n",
+    "print(\"=\" * 50)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.11.15"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}