jingyi0000/VLM_survey: a paper index for vision-language models on vision tasks
Collection of AWESOME vision-language models for vision tasks
At a glance
- What is it?
- The repository behind the TPAMI survey Vision-Language Models for Vision Tasks holds a curated paper list and a short pull request format. It is a reading list and a citation target, not a library, and it is worth adopting only if that is what you need.
- Who is it for?
- Adopt this repository if you need a citation anchor for VLM work on recognition tasks and a paper list that accepts pull requests. Do not adopt it if you expect runnable code, benchmarks or a maintained library; the top level holds only README.md and images/.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 2 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 18, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What jingyi0000/VLM_survey actually is
This repository is the companion page for a survey paper, not a software project. The README states it is the repository of Vision Language Models for Vision Tasks: a Survey, described as a systematic survey of VLM studies in various visual recognition tasks including image classification, object detection and semantic segmentation. The paper itself is linked at arxiv.org/abs/2304.00685 and is published in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2024. The README also notes the paper was selected into the TPAMI Top 50 Popular Paper List.
The audience is narrow and specific. If you are a researcher or an engineer who needs to place a new method against prior work in prompt tuning, pre-training or transfer learning for VLMs, the list is organized exactly along those lines, with sections such as VLM Pre-training Methods and VLM Transfer Learning Methods. If you want to run something, this is the wrong repository. The top level contains README.md and images/, and nothing else.
How the paper list is organized and updated
The mechanism is a Markdown table row. The README gives the submission format explicitly, and it is the only structured data in the project:
|[Title](Paper Link)|Conference|[Code/Project](Code/Project link)|Each entry carries a venue tag in square brackets, such as [CVPR 2025] or [NeurIPS 2024], a title linked to the paper, and where available a second link to code. The README groups entries under headings like VLMs and Synthetic Data, VLM Pre-training Methods and VLM Transfer Learning Methods, and carries a dated news line, Last update on 2025/10/14, plus a note that the arXiv version of the survey is planned to be updated.
The contribution path is a fork, an edit to README.md, and a pull request. There is no schema validation, no CI described in the README, and no generated index. That means the quality of any given row depends on the submitter and the maintainers, and a broken code link is not caught by tooling. For a paper list this is normal. It is still worth knowing before you treat any row as verified.
Getting the list: clone, read, search
There are no install steps because there is nothing to install. The README points to the paper on arXiv and to the repository itself. The practical first use is to clone the README and search it locally, which is faster than scrolling the rendered page and gives you the raw Markdown rows.
git clone https://github.com/jingyi0000/VLM_survey.git
cd VLM_survey
grep -n "CVPR 2025" README.mdAfter the clone you should see README.md and an images directory at the top level. The grep prints every row tagged with that venue, one per line, in the order they appear in the file. Swap the pattern for NeurIPS 2024, ICCV 2025 or a topic word to narrow the list.
To contribute, the README describes the process as forking the project, adding the Title, Paper link, Conference and Project/Code link in README.md using the row format shown above, and submitting the pull request. The README also states that papers missing from the repository can be reported by contact or by opening an issue.
Where this repository stops being useful
The first limitation is scope. The survey covers visual recognition tasks: classification, detection, segmentation. If your work is on spatial reasoning, 3D scenes or medical imaging pipelines, the README does not claim to cover those areas, and the section headings it does list are pre-training, transfer learning and synthetic data. You will be reading a list that was not built for your question.
The second is that a paper list is not a benchmark. The README gives titles, venues and links. It does not report per-method numbers, and there is no evaluation table in the repository. If you need to compare two prompt-tuning methods on a specific dataset, this will tell you the papers exist and nothing more.
The third is freshness. The README carries a dated update line and a note that the arXiv version is planned for update. A curated Markdown file depends on maintainers and contributors keeping up, and nothing in the repository enforces that. Treat any row as a pointer to a paper, not as a statement that the linked code still runs.
Alternatives: paper list versus code base versus survey site
The closest alternative in kind is another curated paper list. The README itself points to one: Awesome-Agentic-MLLMs, a collection on agentic MLLMs, with its own arXiv entry. The difference is subject matter rather than approach. Both are Markdown indexes maintained by pull request, so choosing between them is choosing which literature you care about, not which tooling is better.
A different kind of alternative is a code-first project such as CLIP, which the repository lists as a topic. CLIP ships weights and an API you can call; this repository ships a reading list. The trade-off is direct: CLIP answers what a model outputs on your image, and this repository answers which papers exist and where they were published. Teams often need both, but they are not substitutes, and treating the survey list as an implementation shortcut will not work.
A third option is to read the survey paper itself and skip the repository. The paper is on arXiv and in TPAMI; the repository is the index around it. If you only need the taxonomy and the argument, the paper is the artifact. The repository earns its place when you want to track new entries over time or add your own.
Maintenance, licence and what to verify
The last push to the default branch was on 2026-09-16, so the repository is not dormant. That does not make it actively developed in the software sense; there is no code to develop. What moves is the README, through pull requests and maintainer edits, and the README's own update line is the honest signal of when the list was last touched.
The repository does not state a licence. The README shows an Awesome badge and a PRs-welcome badge, and asks readers to submit pull requests or make contact, but no licence file appears at the top level alongside README.md and images/. That matters if you plan to copy the list into your own documentation or reuse the images directory, which contains an overview figure referenced by the README. Absent a stated licence, assume nothing and ask the maintainers before redistributing. This is a description of what the repository does and does not say, not legal advice.
Upgrade cost is close to zero in the software sense and non-zero in the attention sense. There are no dependencies to bump. There is a list to re-read, and the only way to know what changed is to diff README.md against your last clone.
Editorial conclusion
Adopt this repository if you need a citation anchor for VLM work on recognition tasks and a paper list that accepts pull requests. Do not adopt it if you expect runnable code, benchmarks or a maintained library; the top level holds only README.md and images/. Before relying on it, open the README and confirm which conference entries you need, because the list is updated by hand and the last push was on 2026-09-16.
Frequently asked questions
What does VLM stand for in jingyi0000/VLM_survey?
VLM stands for vision-language model. The repository's README describes the survey as covering VLM studies in visual recognition tasks such as image classification, object detection and semantic segmentation.
Is VLM better than OCR?
The README does not compare VLMs with OCR, so it cannot answer this. It scopes the survey to visual recognition tasks including image classification, object detection and semantic segmentation, and gives no accuracy comparison against OCR systems.
Is VLM better than LLM?
The README does not make this comparison. It describes VLMs as models studied for visual recognition tasks and does not evaluate them against text-only large language models.
Can you give me an example of a Vision Language Model (VLM) from jingyi0000/VLM_survey?
The README lists named methods with their venues, including PLIP for person representation learning at NeurIPS 2024, LoTLIP for long text understanding at NeurIPS 2024, and DPC for tuning vision-language models at CVPR 2025. Each entry links to its paper and, where available, its code.
Community notes