Dataset · tiyunes
SVG primitive text-query annotations (SAM3 v2)
tiyunes/svg-primitive-labels-sam3-v2
Nearly 2M text query and vector graphic element annotation pairs automatically generated using SAM3.
What's inside
Contains 1,926,980 pairs linking English text queries with primitive annotations inside vector graphics based on 382,631 processed SVG files. Data was automatically obtained as pseudo-labels by the SAM3 model without subsequent Gemma filtering, so inaccuracies and extra highlights are possible. This is a standalone second-version extension that excludes the first part.
What you can do with it
- Training models for segmentation and text-based element retrieval within vector graphics
- Training multimodal models for SVG code structure understanding
- Pretraining with subsequent cleaning on stricter data
- Tasks
- Other
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 26.5K
- Likes
- 0
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("tiyunes/svg-primitive-labels-sam3-v2")