Worklog

---

info

This page is solely for recording our work activities.

---

2024

September

• MRZScanner: Project completed and made open-source. 🎉 🎉 🎉
• TextDetector: Continued development, following progress made in March.
• Came across a beautifully designed website and had to note it down:
  • Hello Algorithm
• Wrote paper reviews, totaling approximately 100 papers.

August

• MRZScanner: Deployment testing and rework.
• Updated @docusaurus/core@3.5.2 and found out it's not backward compatible...
  • Spent time updating the problematic code sections.
• Investigated OpenCV dependency issues and discovered we weren’t alone:
  • 修复 OpenCV 依赖错误的小工具：OpenCV Fixer
  • Open-source project: soulteary/opencv-fixer
  • Thanks to 蘇洋博客 for sharing and saving us a lot of time.
• Wrote paper reviews, with approximately 90 papers reviewed so far.

July

• Wrote paper reviews, with around 80 papers in total so far.
• MRZScanner: Began development.

June

• AutoTraderX: Completed the API integration for Yuanta Securities and made it open-source. 🎉 🎉 🎉
• Ran out of funds for OpenAI services, so we suspended the daily news push from GmailSummary.
• Continued writing paper reviews, totaling 50 papers by this point.

May

• Finished developing the Text Recognizer model.
  • Final evaluation results were promising, but we think it’s still an "overfitted model pretending not to be overfitted." (???)
  • Since it doesn’t meet our ideal standards yet, we’ve decided not to release it for now.
• Explored Docusaurus' Search feature, tested and integrated Algolia search service.
  • Thanks to WeiWei for the tutorial:
    • [docusaurus] 在 Docusaurus 中使用 Algolia 實作搜尋功能
• Continued working on the Text Recognizer model, adjusting parameters and training.
• AutoTraderX: Development started.

April

• Learned how to configure CSS styles to tweak the blog’s appearance.
  • Thanks to 朝八晚八 for the helpful article:
    • 部落格首頁
• TextRecognizer: Continued development from WordCanvas and made further progress on the text recognition project.
• GmailSummary: Modified functionality to push daily news to the tech documentation page.
• Completed technical documentation for all ongoing projects.
• Explored Docusaurus’ i18n functionality and started writing English documentation.
• Investigated Docusaurus’ documentation features and began migrating content from GitHub to the platform.
• WordCanvas: Project completed and made open-source. 🎉 🎉 🎉
• Renamed the blog GitHub project to website for accuracy.

March

One day, we found that the Google Drive download feature broke—what was once accessible through gen_download_cmd became a garbled mess of HTML. 👻 👻 👻

After considering several options...

We decided to use NextCloud to set up a private cloud for storing data and updated our previous download links accordingly.

• GmailSummary: Completed development and made it open-source. 🎉 🎉 🎉
• DocClassifier: Discovered that stacking multiple normalization layers significantly improved model performance (a surprising discovery...).
• TextRecognizer: Early-stage project planning.
• WordCanvas: Development started.
• TextDetector: Ran into several issues and decided to put it on hold for now.

February

• TextDetector: Collected public datasets.
• DocClassifier: Introduced CLIP into the model and applied knowledge distillation with excellent results!
• Explored Docusaurus’ commenting functionality and integrated giscus for comments.
  • Thanks to 不務正業的架構師 for the insightful guide:
    • 在文件庫和部落格的文章下方加上 giscus 留言區

January

• TextDetector: Early-stage project planning.
• DocClassifier: Project completed and made open-source. 🎉 🎉 🎉

2023

December

• DocClassifier: Development started.
• DocAligner: Completed development and made it open-source. 🎉 🎉 🎉
• Blog: Discovered Meta’s interesting open-source project Docusaurus. It provides a simple way to build a static website using Markdown for content creation, so I decided to use it to write a blog.
• Abandoned and deleted the WordPress-built website, migrating all content to the GitHub blog project.

November

• DocClassifier: Early-stage project planning.
• DocsaidKit: Completed development and made it open-source. 🎉 🎉 🎉
• Wrote paper reviews, totaling 20 papers.

October

• WordCanvas: Early-stage project planning.
• DocGenerator: Completed phase two of development, splitting the text synthesis module into the WordCanvas project.

September

• DocAligner: Development started.
• DocGenerator: Phase one of development completed.
• Wrote paper reviews, totaling 5 papers.

August

• DocAligner: Early-stage project planning.
• DocsaidKit: Organized commonly used tools and started development.
• Explored WordPress functionality, experimented with building a personal blog.
  • Thanks to 諾特斯網站 for the generous knowledge sharing.
• Created a DOCSAID GitHub account and started planning various projects.