Clipvg

Environment

• CUDA 11.3
• Python 3.6
• pip install -r requirements.txt
• Install diffvg:

git clone git@github.com:BachiLi/diffvg.git --recursive
cd diffvg
python setup.py build

CLIPVG Optimization

• Default Setting

python main.py \
--svg images/pitt.svg \
--prompts "Joker, Heath Ledger"

• ROI Prompts

# The first prompt is always defined for the whole image, followed by the extra ROI prompts.
# The extra ROIs are defined by x1, y1, w1, h1, x2, y2, w2, h2, ...
python main.py \
--svg images/horseriding.svg \
--prompts "Astronaut riding a metal horse" "Astronaut" "A metal horse" \
--extra_rois 185 84 185 316 20 175 480 325

• Shape-only manipulation

python main.py \
--svg images/athlete.svg \
--prompts "Stephen Curry" \
--shape_lr 0.4 \
--color_lr 0.0

• Color-only manipulation

python main.py \
--svg images/tajmahal.svg \
--prompts "Taj Mahal under the moon light" \
--shape_lr 0.0 \
--color_lr 0.02

Results

• Find the results in ./output/<image_name>/<prompt>_<exp_name>
• The default exp_name is "exp"

Configs

• Check the configs in ./config/configs.py

Acknowledgement

• Our code is based on diffvg and CLIP