R1l3y-w/im2vec-svg-emoji

im2vec-svg-emoji

Rendered (SVG, PNG) pairs for training Im2Vec, a raster-logo-to-SVG model. This is a pre-rendered version of starvector/svg-emoji: each .svg file is paired with a same-stem .png rendered at 256x256 via cairosvg, matching this project's training pipeline (im2vec/data/render.py).

Why this dataset

The project's original training data, FIGR-8, turned out to contain zero color information — every sampled training SVG relies on the SVG default fill (black), so a model trained on it only ever learns to output black shapes. svg-emoji is full-color and was added as a fix.

Structure

train/svg/*.svg   train/png/*.png    (8708 pairs)
valid/svg/*.svg   valid/png/*.png    (667 pairs)
test/svg/*.svg    test/png/*.png     (668 pairs)

SVGs and PNGs are split into separate svg//png/ subfolders per split (Hugging Face caps a single directory at 10,000 files, and each split has more than 10,000 combined). .svg/.png pairs share the same filename stem — merge them into one directory to use directly with im2vec.dataset.load_manifest, e.g.:

mkdir -p data/svg-emoji/train
cp train/svg/*.svg train/png/*.png data/svg-emoji/train/

License

svg-emoji combines several open emoji sets, each under its own license — check each file's origin suffix in its filename:

• —*-twemoji.svg — Twemoji, CC-BY 4.0
• —*-noto-emoji.svg — Noto Emoji, Apache-2.0 / OFL
• —*-openmoji.svg — OpenMoji, CC BY-SA 4.0 (share-alike)

Any redistribution must honor the license of each source file it includes, in particular OpenMoji's share-alike requirement.

Reproduce

python -m im2vec.data.prepare --dataset svg-emoji --split train --out data/svg-emoji/train
python -m im2vec.data.prepare --dataset svg-emoji --split valid --out data/svg-emoji/valid
python -m im2vec.data.prepare --dataset svg-emoji --split test  --out data/svg-emoji/test