fei-aiart/courses: A Chinese-Language Slide Archive for Image Processing, Deep Learning and Computer Vision

What the repository actually contains

This is not a library and not a framework. It is a set of lecture slides for four Chinese-language courses: digital image processing, deep learning, computer vision, and machine learning with deep learning. The README lists seven decks in a table with a course name, a link, an update year and a short note. The newest entry is a deep learning deck updated 2025.10, described as a 2-4 hour crash version that runs from MLP to MLLM. The oldest entries go back to 2020 and 2021. There is a companion set of short essays on study habits, research process, technical writing and graduate school interviews, stored under a blogs path.

The intended audience is clear from the usage notes: the README says non-commercial and university teaching use is free. That sentence is the whole licence. The repository metadata reports no licence identifier, and there is no LICENSE file mentioned anywhere in the supplied material. A reader who needs to redistribute these slides, embed them in a paid course, or ship them inside a product has no written permission to point at. Treat that as an open question to resolve with the author, not as an implied yes.

The link table is the delivery mechanism, and it is uneven

There is no package, no build step and no install command. Distribution happens through links in the README table, and those links are of three different kinds. The deep learning 2025.10 deck and the machine learning deck point at files inside the repository itself, one as a URL-encoded PDF path and one as a relative markdown file. The digital image processing, deep learning 2024 and computer vision decks point at Google Drive. The computer vision and computer vision basics decks point at Baidu Pan, and the image processing deck also has a Jianguoyun mirror.

That mix has practical consequences. Google Drive and Baidu Pan links can expire, hit quota limits or be regionally unreachable, and only two of the seven entries are stored in the repository where they cannot rot. Two Baidu links carry extraction codes in backticks (`wqkc`, `3zb7`) and one Jianguoyun entry carries `mdykW9`, which means a reader has to copy a code before the download will start. If you are assembling a course for a semester, mirror the cloud-hosted decks locally on day one. The repository's own git history only protects the two in-repo files.

Fonts, master slides and the editing workflow the README prescribes

The usage section gives three instructions, and they describe a PowerPoint-centric workflow rather than a code-centric one. First, install Alibaba PuHuiTi (阿里巴巴普惠体), linked from iconfont.cn, because without it the typography breaks. Second, to change the logo, open the PPT master view (母版视图) and replace the school badge and related information. Third, the licence note about non-commercial university teaching.

The master-view instruction is the interesting one. It tells you the decks are templates, not fixed documents: a lecturer is expected to swap the institutional identity and reuse the body. That is a reasonable design for teaching material and it is also why the licence matters more than it would for a blog post. Once you have replaced a logo and started editing, you are distributing a derivative. The README's one-line permission covers non-commercial and university teaching; it does not address modification, attribution wording, or redistribution outside a classroom. The README also opens with a note that the material references many online resources and invites corrections for infringement or errors, which is an acknowledgement that some figures may not be the author's own. If you plan to publish a derived deck, that is the second thing to audit after the font.

Where the material is thin and where it stands on its own

The README's own notes are the most honest part of the listing. The computer vision deck from 2021 is flagged as best used alongside Stanford CS231n, which means it is a companion rather than a complete course. The computer vision basics deck is described as leaning toward traditional methods with little deep learning. The digital image processing deck from 2022 is tied to Gonzalez and Woods, Digital Image Processing, 4th edition. The 2024 deep learning deck cites MIT 6.S191, CS231n and a Chinese Academy of Sciences AICS course, and recommends the PaddlePaddle 'dandelion book' (nndl.github.io) as a textbook.

Read that as a dependency graph. Several decks assume an external syllabus is doing the heavy lifting, so a self-directed learner who downloads only the slides gets slides without the reading list, the assignments or the lecture recordings. The 2025.10 deck is the exception in framing: a 2-4 hour crash course from MLP to MLLM is a self-contained shape, and it is the only entry updated this decade's second half. Nothing in the material states how many slides any deck contains, so you cannot judge depth from the README alone. Open the PDF before committing a syllabus to it.

Why this is not a software project, and what that rules out

The primary language is listed as HTML, which is misleading in a useful way: the repository is a content archive whose web-facing parts are markdown and PDF links. There are no releases, no dependency manifest, no test suite, no API and no configuration keys. If you arrived expecting to `pip install` something, or to find a Python implementation of the algorithms the slides describe, you will not find it here. There is nothing to run.

That is the main failure mode: mistaking a teaching archive for a codebase. A second failure mode is version drift. The decks span 2020 to 2025, and the deep learning field moves faster than a slide deck does. A 2021 computer vision deck that already defers to CS231n is teaching the state of the art as of its writing, not as of today. The README gives update years precisely so you can see this, which is more than many course repositories do, but it does not give changelogs, so you cannot tell what changed between the 2024 and 2025.10 deep learning decks.

What to use instead, and the actual difference

If you want slides plus a runnable codebase, the d2l book (Dive into Deep Learning) is the natural alternative, and the difference is structural rather than a matter of quality. d2l interleaves prose, math and executable notebooks in one repository under an open licence, so the code you read is the code you run. This repository gives you slides only, in Chinese, with no execution environment. If you want an English-language course with assignments and video, CS231n is the alternative the README itself points to for computer vision, and it comes with a structured syllabus this repository does not attempt. If your need is a textbook on image processing, Gonzalez and Woods is what the 2022 deck is built on, and buying the book gets you the exercises and figures with clear rights. The honest framing: fei-aiart/courses is a starting point for a lecturer who wants to teach in Chinese and edit rather than author from scratch. It is not a substitute for any of the above if you need code, English, or a licence you can rely on.

Maintenance cost and what to verify before you adopt it

Maintenance here is not dependency upgrades. It is link rot, font availability and licence scope. The last push to the repository was 2026-09-07, and the newest deck is dated 2025.10, so the project is alive but not on a release cadence; there are no releases to track. The recurring cost is checking that the Google Drive, Baidu Pan and Jianguoyun links still resolve, and that the extraction codes still work, every time you hand the list to a new cohort. The font dependency is a one-time install per machine, and it is a hard requirement per the README, not a suggestion.

The licence is the item to resolve first. A missing licence file plus a README sentence granting free use for non-commercial and university teaching is a narrow permission, and it does not say what happens to modified versions or to online publication of a derived deck. That is a question for the author, not something to infer. The repository's questionnaire link suggests the author is actively interested in who uses the material, which makes asking straightforward. Verify three things in order: whether your intended use falls inside 'non-commercial and university teaching', whether the specific deck you want has third-party figures you would need to replace, and whether the cloud link for that deck is still live. The two decks stored in the repository itself are the ones you can archive with confidence.

Editorial conclusion

Adopt this if you teach or study digital image processing, computer vision or deep learning in Chinese and need ready-made decks you can edit in PowerPoint; the README permits non-commercial use in university teaching. Do not adopt it if you need English material, a permissively licensed asset you can redistribute inside a product, or anything installable. Before using a deck in a course, open the PPT master view and confirm the school logo and any third-party figures, install Alibaba PuHuiTi so the layout does not reflow, and treat the missing LICENSE file as the blocking question rather than a formality.

Official sources

1. fei-aiart/courses on GitHub
2. Issues
3. Project website
4. README