Skip to content
View u-wlkjyy's full-sized avatar

Block or report u-wlkjyy

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
u-wlkjyy/README.md
typing svg

About

I'm wlkjyy, a software engineering researcher and full-stack engineer. My work sits at the intersection of computer vision and large language models, with a focus on making foundation models practical — adapting them with few parameters, little data, and real deployment constraints.

Beyond research, I design and build the systems around it: I take ideas from prototypes to production, including the model pipeline, the services, and the infrastructure they run on.

Research Interests

  • Parameter-Efficient Fine-Tuning (PEFT) — lightweight adaptation of large foundation models (e.g., SAM 3) via adapters and frozen backbones
  • Few-Shot Object Detection — transferring general-purpose promptable models to fixed-class detectors under minimal supervision
  • Sequence Recognition in the Wild — deep learning for variable-length text recognition, using captchas as a benchmark
  • AI Systems Engineering — tooling, acceleration, and deployment for machine learning workloads

Selected Work

Project Summary
sam3-box-adapter Parameter-efficient adaptation of SAM 3 for few-shot, box-only detection: frozen ViT backbone, residual adapters, and a lightweight detection head — no text prompts or mask supervision required.
DeepCaptcha-Solver A production-grade deep learning pipeline for recognizing variable-length character sequences.
siam-autolabel Training data auto-labeled by a multimodal LLM (Gemini), distilled into a siamese network for grid-captcha verification.
JCo A minimal Java concurrency library exploring JDK 21 virtual threads and structured concurrency to bring coroutine-style semantics to the JVM.
gd-opencv-php An OpenCV-style image-processing API implemented in pure PHP on top of GD.
ollama-fast An accelerator for downloading and distributing LLM runtimes and models (Ollama).
safedel A high-performance secure file deletion utility.
pathutil A Go CLI for managing the Windows PATH environment variable via the registry.

Open Source

Occasional contributor to the broader ecosystem — maintaining forks of Goravel route drivers (gin / fiber), the RuoYi-Vue-Ultra admin console, and winget package manifests.

Statistics

stats top languages
activity graph

© 2026 wlkjyy · github.com/u-wlkjyy · views

Popular repositories Loading

  1. mumu-python-api mumu-python-api Public

    Mumu模拟器Python API

    Python 103 29

  2. siam-autolabel siam-autolabel Public

    Gemini自标注与孪生网络驱动的九宫格验证码识别

    Python 18 3

  3. ollama-fast ollama-fast Public

    ollama模型下载加速工具

    Go 11 2

  4. uniapp-android-builder-docker uniapp-android-builder-docker Public

    Uniapp安卓离线打包的docker镜像

    HTML 2

  5. laravel-project-toolkit laravel-project-toolkit Public

    快速创建laravel项目的工具

    Go 1

  6. JCo JCo Public

    JCo 是一个极简的 Java 并发库,它利用 JDK 21 原生的 虚拟线程 (Virtual Threads) 和 结构化并发 (Structured Concurrency),在 Java 中完美复刻了 Kotlin 协程 的优雅开发体验。

    Java 1