ParkerLabel: CPU-only image annotation with MobileSAM

Parker Lyu — Fri, 09 Oct 2026 06:16:26 +0000

I open-sourced ParkerLabel, a desktop tool for object detection and instance segmentation annotation. It runs locally on the CPU. No GPU or image upload is needed.

In AI mode, positive and negative clicks guide MobileSAM to generate an object mask. You can correct it manually, assign a category, and save the annotation. The interface has separate image, mask, and overlay views.

The inference path uses ONNX Runtime. The image embedding is saved beside the image and reused. Each instance keeps its own mask, category, bounding box, and area in a per-image JSON file. Masks use uncompressed COCO RLE; these files are not a complete COCO dataset export.

There are portable packages for macOS arm64 and Windows x64. Extract the folder to a writable location and run the app; no Python environment is needed. The UI supports nine languages, including English and Chinese.

Source · Download

If you work with image annotations, give it a try. Bug reports and workflow feedback are welcome.

Learn TensorRT: a C++17 path from YOLOv8 to async inference

Parker Lyu — Fri, 09 Oct 2026 05:55:21 +0000

I open-sourced Learn TensorRT, a hands-on course for developers moving from PyTorch models to C++ deployment. The baseline is TensorRT 10.14, CUDA 13.0 and C++17, in a pinned NVIDIA development container.

The core path uses YOLOv8 to connect three parts of deployment:

Correctness: export to ONNX, inspect outputs with Polygraphy, then implement preprocessing, TensorRT inference and postprocessing in C++.
Optimization: compare FP32/FP16/INT8, use explicit Q/DQ quantization, and profile with Nsight.
Pipeline behavior: add bounded queues, dynamic batching and asynchronous CUDA streams, then measure latency, throughput and overload behavior.

A useful starting point is the single-image C++ pipeline in lesson 11. Follow its prerequisites, check the detections, and establish correctness before changing precision or adding concurrency. Later reports separate engine timing from application and pipeline measurements.

The code emphasizes RAII, explicit resource ownership and target-based CMake. Lessons have build/run instructions and reporting checkpoints. Ubuntu with an NVIDIA GPU is the reference setup; C++ and CMake basics are expected. Engines and performance results must be regenerated for your environment.

Electives include plugins, Triton, DeepStream and Jetson/DLA. Some elective runtime acceptance is still pending; the coverage matrix records those limits.

Repository and learning roadmap — MIT licensed, with English, Chinese, Japanese and Korean READMEs.

If you're learning TensorRT deployment, try the core path. Feedback on unclear steps or reproducibility problems is welcome.

AI assistance was used to draft this introduction.

DEV Community: Parker Lyu

ParkerLabel: CPU-only image annotation with MobileSAM

Learn TensorRT: a C++17 path from YOLOv8 to async inference