HaoyuanYang-2023/ImagineFSL

Official implementation of "ImagineFSL: Self-Supervised Pretraining Matters on Imagined Base Set for VLM-based Few-shot Learning" [CVPR 2025 Highlight]

17
/ 100
Experimental

ImagineFSL helps computer vision researchers and practitioners efficiently train image classification models, especially when only a few real-world examples are available. It takes descriptive text (captions) and a small set of real images as input, then generates a large synthetic dataset and uses it to pre-train a vision model. The output is a robust image classification model capable of accurately categorizing images even with limited real-world data.

No commits in the last 6 months.

Use this if you need to build accurate image classification models for new categories or tasks but have very few real training images and want to leverage large language models and image generation for data augmentation.

Not ideal if you already have extensive real-world datasets for your classification task or if your primary goal is not image classification.

computer-vision image-classification data-synthesis model-training few-shot-learning
No License Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 7 / 25
Maturity 8 / 25
Community 0 / 25

How are scores calculated?

Stars

26

Forks

Language

Python

License

Last pushed

Sep 01, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/HaoyuanYang-2023/ImagineFSL"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.