astra-vision/PODA

[ICCV 2023] New framework: Domain adaptation using a single prompt. Main contribution: Prompt-driven Instance Normalization (PIN)

39
/ 100
Emerging

PODA helps computer vision engineers adapt image recognition models to new, unexperienced environments using only a short text description. For example, if you trained a model on daytime images, PODA lets you adapt it to night-time images by simply describing "driving at night." It takes your existing image recognition model and a text description, then outputs an improved model that performs better in the described conditions.

124 stars. No commits in the last 6 months.

Use this if you need to quickly adapt an image recognition or object detection model to a new visual domain without collecting and labeling new training data for that specific domain.

Not ideal if you have ample labeled data for your target domain or if your domain shift is too complex to be described by a single text prompt.

computer-vision image-segmentation object-detection model-adaptation autonomous-driving
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 13 / 25

How are scores calculated?

Stars

124

Forks

13

Language

Python

License

Apache-2.0

Last pushed

Mar 15, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/astra-vision/PODA"

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