- 🎯 方向：目标检测、多目标跟踪、位姿解算与模型量化 / 端侧部署。
- 📖 目前在做「翻页检测」：找书（书 + 手）→ 目标跟踪 → 翻页判定，跑通从数据到板端的完整链路。
- 🧪 也在把自训的 YOLOv8 + DSConv 轻量骨干（4 类，1.99 M 参数）从 PyTorch → ONNX → RKNN int8 量化，部署到 RV1106（Luckfox Pico） 板端。
- 🤖 写过 RoboMaster 自瞄视觉：灯条传统检测与神经网络检测并行，配 PnP 解算、EKF 预测与 ByteTrack 跟踪。
- 🧱 习惯把经典结构自己手写一遍 —— 比如从零实现 YOLOv8（Anchor-Free + DFL + Task-Aligned Assigner），而不是只调库。
- 🌱 正在补：C++ 工程化、CUDA / 算子优化、多传感器标定与融合。
| 项目 | 说明 | 技术栈 |
|---|---|---|
| yolov8 | 从零手写 YOLOv8：模型结构与配置、DFL 边框回归、Task-Aligned 标签分配、损失与训练/推理流程 | PyTorch Python |
| rm | RoboMaster 自瞄视觉（Mac 版）：传统灯条检测 + YOLO 检测、PnP 解算、EKF、BYTETracker、决策与配置模块 | C++ OpenCV Eigen CMake |
| bishe | YOLO 检测 + ROS 2 集成：COCO→YOLO 数据转换、模型训练、pt→ONNX 导出、检测发布/订阅与 RViz 可视化 | Python ROS 2 RViz |
| Armor_detect | 装甲板检测早期版本：灯条配对、装甲板匹配、卡尔曼预测与 PnP 位姿解算 | C++ OpenCV |
🧰 技能栈 / Tech Stack
| 分类 | 内容 |
|---|---|
| 语言 | Python、C++ |
| 深度学习 | PyTorch、Ultralytics YOLO、ONNX、RKNN（Rockchip NPU） |
| 视觉 | OpenCV、PnP / 相机标定、卡尔曼滤波、EKF、ByteTrack / SORT |
| 工程 | CMake、Linux、Git、ROS 2、RViz、Docker |
欢迎交流目标检测、跟踪与端侧部署相关的问题，也欢迎 Issue / PR。
Hi, I'm Marking (@cpolarc) — a computer-vision engineer focused on object detection, multi-object tracking, pose estimation, and embedded AI deployment.
- 🔭 Currently building: a page-turn detection pipeline (find the book & hands → track → detect the page flip), from dataset to on-device inference.
- 🧪 Also working on: quantizing a self-trained YOLOv8 + DSConv model (4 classes, 1.99 M params) through PyTorch → ONNX → RKNN int8, deployed on an RV1106 (Luckfox Pico) board.
- 🤖 Past work: RoboMaster auto-aim vision — traditional light-bar detection running alongside a neural detector, with PnP pose solving, EKF prediction and ByteTrack.
- 🧱 I like re-implementing classic architectures from scratch (e.g. YOLOv8 with Anchor-Free heads, DFL regression and a Task-Aligned Assigner) instead of only calling libraries.
Selected projects
| Project | What it is | Stack |
|---|---|---|
| yolov8 | YOLOv8 written from scratch: model/config, DFL box regression, Task-Aligned label assignment, losses, training & inference | PyTorch |
| rm | RoboMaster auto-aim vision: light-bar & YOLO detection, PnP solving, EKF, BYTETracker, decision module | C++ OpenCV Eigen |
| bishe | YOLO detection + ROS 2 integration: dataset conversion, training, pt→ONNX export, detection pub/sub and RViz visualization | Python ROS 2 |
| Armor_detect | Earlier armor-detection version: light-bar pairing, armor matching, Kalman prediction, PnP | C++ OpenCV |
📫 Reach me at 3550959735@qq.com — happy to talk about detection, tracking or edge deployment.