roboflow · yashdivya5 · May 21, 2026
diff --git a/notebooks/train-yolo26-pose-estimation-on-custom-dataset.ipynb b/notebooks/train-yolo26-pose-estimation-on-custom-dataset.ipynb
@@ -0,0 +1,359 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+  "colab": {
+   "provenance": [],
+   "gpuType": "T4"
+  },
+  "kernelspec": {
+   "name": "python3",
+   "display_name": "Python 3"
+  },
+  "language_info": {
+   "name": "python"
+  },
+  "accelerator": "GPU"
+ },
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "<div align=\"center\">\n",
+    "  <a href=\"https://roboflow.com\" target=\"_blank\">\n",
+    "    <img\n",
+    "      width=\"100%\"\n",
+    "      src=\"https://media.roboflow.com/notebooks/template/bannerformats-dark.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949716562\"\n",
+    "    >\n",
+    "  </a>\n",
+    "</div>\n",
+    "\n",
+    "# Fine-Tune YOLO26 on Pose Estimation Dataset\n",
+    "\n",
+    "---\n",
+    "\n",
+    "[![Roboflow](https://raw.githubusercontent.com/roboflow-ai/notebooks/main/assets/badges/roboflow-blogpost.svg)](https://roboflow.com)\n",
+    "[![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://youtube.com/roboflow)\n",
+    "[![GitHub](https://badges.aleen42.com/src/github.svg)](https://github.com/ultralytics/ultralytics)\n",
+    "\n",
+    "YOLO26 introduces **NMS-free end-to-end inference** and the **MuSGD optimizer**, making it faster and more accurate than previous YOLO versions. This notebook shows you how to fine-tune YOLO26 on a custom pose estimation dataset using [Roboflow](https://roboflow.com) for dataset management.\n",
+    "\n",
+    "**What you'll learn:**\n",
+    "- How to prepare a pose estimation dataset with Roboflow\n",
+    "- How to fine-tune YOLO26 on a custom keypoint dataset\n",
+    "- How to run inference and visualize keypoints\n",
+    "- How to export the model for deployment\n",
+    "\n",
+    "**Before you start**, make sure you have access to a GPU. You can use a **free T4 GPU** by navigating to `Runtime` → `Change runtime type` → `T4 GPU`."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 1: Install Dependencies"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "HOME = os.getcwd()\n",
+    "print(\"HOME:\", HOME)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%pip install ultralytics supervision roboflow -q\n",
+    "\n",
+    "import ultralytics\n",
+    "ultralytics.checks()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 2: Load Pretrained YOLO26 Pose Model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from ultralytics import YOLO\n",
+    "\n",
+    "# Load YOLO26 nano pose model pretrained on COCO\n",
+    "model = YOLO(\"yolo26n-pose.pt\")\n",
+    "print(\"Model loaded successfully!\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 3: Download Dataset from Roboflow\n",
+    "\n",
+    "We'll use a publicly available pose estimation dataset from Roboflow Universe. You can replace this with your own dataset.\n",
+    "\n",
+    "**To use your own dataset:**\n",
+    "1. Go to [Roboflow Universe](https://universe.roboflow.com)\n",
+    "2. Find or upload your pose estimation dataset\n",
+    "3. Export in **YOLOv8 Pose** format\n",
+    "4. Replace the snippet below with your own export code"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from roboflow import Roboflow\n",
+    "\n",
+    "rf = Roboflow(api_key=\"YOUR_ROBOFLOW_API_KEY\")  # Replace with your Roboflow API key\n",
+    "project = rf.workspace(\"roboflow-jvuqo\").project(\"football-players-detection-3zvbc\")\n",
+    "version = project.version(1)\n",
+    "dataset = version.download(\"yolov8-pose\")\n",
+    "\n",
+    "print(f\"Dataset downloaded to: {dataset.location}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 4: Fine-Tune YOLO26 on Pose Estimation Dataset"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from ultralytics import YOLO\n",
+    "\n",
+    "model = YOLO(\"yolo26n-pose.pt\")\n",
+    "\n",
+    "results = model.train(\n",
+    "    data=f\"{dataset.location}/data.yaml\",\n",
+    "    epochs=100,\n",
+    "    imgsz=640,\n",
+    "    batch=16,\n",
+    "    device=0,\n",
+    "    project=\"pose_training\",\n",
+    "    name=\"yolo26n_pose\",\n",
+    "    exist_ok=True,\n",
+    "    patience=20,\n",
+    "    save=True,\n",
+    "    plots=True,\n",
+    ")\n",
+    "\n",
+    "print(f\"\\nTraining complete!\")\n",
+    "print(f\"Best model saved to: {results.save_dir}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 5: Evaluate the Model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from ultralytics import YOLO\n",
+    "\n",
+    "# Load best trained model\n",
+    "model = YOLO(\"pose_training/yolo26n_pose/weights/best.pt\")\n",
+    "\n",
+    "# Run validation\n",
+    "metrics = model.val(\n",
+    "    data=f\"{dataset.location}/data.yaml\",\n",
+    "    device=0,\n",
+    ")\n",
+    "\n",
+    "print(f\"Pose mAP50:   {metrics.pose.map50:.4f}\")\n",
+    "print(f\"Pose mAP50-95: {metrics.pose.map:.4f}\")\n",
+    "print(f\"Box mAP50:    {metrics.box.map50:.4f}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 6: Run Inference and Visualize Keypoints"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import glob\n",
+    "import random\n",
+    "from IPython.display import display, Image as IPImage\n",
+    "from ultralytics import YOLO\n",
+    "\n",
+    "model = YOLO(\"pose_training/yolo26n_pose/weights/best.pt\")\n",
+    "\n",
+    "# Pick a random validation image\n",
+    "val_images = glob.glob(f\"{dataset.location}/valid/images/*.jpg\")\n",
+    "test_image = random.choice(val_images)\n",
+    "\n",
+    "results = model.predict(\n",
+    "    source=test_image,\n",
+    "    conf=0.35,\n",
+    "    save=True,\n",
+    "    project=\"pose_inference\",\n",
+    "    name=\"yolo26_pose_results\",\n",
+    "    exist_ok=True,\n",
+    ")\n",
+    "\n",
+    "# Display result\n",
+    "output_image = glob.glob(\"pose_inference/yolo26_pose_results/*.jpg\")[0]\n",
+    "display(IPImage(output_image, width=800))\n",
+    "\n",
+    "# Print keypoint details\n",
+    "for r in results:\n",
+    "    if r.keypoints is not None:\n",
+    "        print(f\"Detected {len(r.keypoints)} person(s)\")\n",
+    "        print(f\"Keypoints shape: {r.keypoints.xy.shape}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 7: Visualize with Supervision"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import supervision as sv\n",
+    "import cv2\n",
+    "from ultralytics import YOLO\n",
+    "from IPython.display import display, Image as IPImage\n",
+    "import tempfile, os\n",
+    "\n",
+    "model = YOLO(\"pose_training/yolo26n_pose/weights/best.pt\")\n",
+    "\n",
+    "image = cv2.imread(test_image)\n",
+    "results = model(image, conf=0.35)[0]\n",
+    "\n",
+    "# Extract detections and keypoints\n",
+    "keypoints = sv.KeyPoints.from_ultralytics(results)\n",
+    "detections = sv.Detections.from_ultralytics(results)\n",
+    "\n",
+    "# Annotate\n",
+    "box_annotator = sv.BoxAnnotator()\n",
+    "edge_annotator = sv.EdgeAnnotator(color=sv.Color.GREEN, thickness=2)\n",
+    "vertex_annotator = sv.VertexAnnotator(color=sv.Color.RED, radius=4)\n",
+    "\n",
+    "annotated = box_annotator.annotate(image.copy(), detections)\n",
+    "annotated = edge_annotator.annotate(annotated, keypoints)\n",
+    "annotated = vertex_annotator.annotate(annotated, keypoints)\n",
+    "\n",
+    "# Save and display\n",
+    "output_path = \"pose_supervision_output.jpg\"\n",
+    "cv2.imwrite(output_path, annotated)\n",
+    "display(IPImage(output_path, width=800))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Step 8: Export for Deployment"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from ultralytics import YOLO\n",
+    "\n",
+    "model = YOLO(\"pose_training/yolo26n_pose/weights/best.pt\")\n",
+    "\n",
+    "# Export to ONNX\n",
+    "model.export(format=\"onnx\", imgsz=640, simplify=True)\n",
+    "print(\"Exported to ONNX\")\n",
+    "\n",
+    "# Export to TFLite (mobile devices)\n",
+    "# model.export(format=\"tflite\", imgsz=640, int8=True)\n",
+    "\n",
+    "# Export to TensorRT (NVIDIA GPUs)\n",
+    "# model.export(format=\"engine\", imgsz=640, half=True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## What's Next?\n",
+    "\n",
+    "- 📖 [YOLO26 Docs](https://docs.ultralytics.com/models/yolo26)\n",
+    "- 🔍 [Roboflow Universe — Pose Datasets](https://universe.roboflow.com/search?q=pose+estimation)\n",
+    "- 💜 [Supervision Docs](https://supervision.roboflow.com)\n",
+    "- 🐛 [Report Issues](https://github.com/roboflow/notebooks/issues)\n",
+    "\n",
+    "<div align=\"center\">\n",
+    "  <div>\n",
+    "    <a href=\"https://youtube.com/roboflow\">\n",
+    "        <img\n",
+    "          src=\"https://media.roboflow.com/notebooks/template/icons/purple/youtube.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949748453\"\n",
+    "          height=\"110\"\n",
+    "        >\n",
+    "    </a>\n",
+    "    <a href=\"https://roboflow.com\">\n",
+    "        <img\n",
+    "          src=\"https://media.roboflow.com/notebooks/template/icons/purple/roboflow-app.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949748453\"\n",
+    "          height=\"110\"\n",
+    "        >\n",
+    "    </a>\n",
+    "    <a href=\"https://www.linkedin.com/company/roboflow-ai/\">\n",
+    "        <img\n",
+    "          src=\"https://media.roboflow.com/notebooks/template/icons/purple/linkedin.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949748453\"\n",
+    "          height=\"110\"\n",
+    "        >\n",
+    "    </a>\n",
+    "    <a href=\"https://roboflow.com/twitter\">\n",
+    "        <img\n",
+    "          src=\"https://media.roboflow.com/notebooks/template/icons/purple/twitter.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949748453\"\n",
+    "          height=\"110\"\n",
+    "        >\n",
+    "    </a>\n",
+    "    <a href=\"https://discord.gg/roboflow\">\n",
+    "        <img\n",
+    "          src=\"https://media.roboflow.com/notebooks/template/icons/purple/discord.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949748453\"\n",
+    "          height=\"110\"\n",
+    "        >\n",
+    "    </a>\n",
+    "  </div>\n",
+    "</div>"
+   ]
+  }
+ ]
+}