A Collaborative, Interactive and Context Aware Drawing Agent for Co-Creative Design
Francisco Ibarrola, Tomas Lawton and Kazjon Grace
You can read the preprint here
To avoid the struggle that installation sometimes entail, we recommend you try CICADA in a google Colab notebook.
Make sure your gcc version is 8 or lower
gcc --version
Make sure you have Anaconda 3 installed. Clone the repo, cd to it and run setup.sh (this can take a while)
git clone https://github.com/fibarrola/cicada.git
cd cicada
source setup.sh
Once installed, it can be run as
$ python3 main.py
The following options can be used to change the default parameters
--svg_path <str> path to svg partial sketch
--prompt <str> what to draw
--neg_prompt <str>
--neg_prompt_2 <str>
--use_neg_prompts <bool>
--normalize_clip <bool>
--num_paths <int> number of strokes to add
--canvas_h <int> canvas height
--canvas_w <int> canvas width
--max_width <int> max px width
--num_iter <int> maximum algorithm iterations
--w_points <float> regularization parameter for Bezier point distance
--w_colors <float> regularization parameter for color differences
--w_widths <float> regularization parameter for width differences
--w_img <float> regularization parameter for image L2 similarity
--w_geo <float> regularization parameter for geometric similarity
--x0 <float> coordinate for drawing area
--x1 <float> coordinate for drawing area
--y0 <float> coordinate for drawing area
--y1 <float> coordinate for drawing area
--num_trials <int> number of times to run the algorithm
--num_augs <int> number of augmentations for computing semantic loss
--prune_ratio <float> ratio of paths to be pruned out
--save_path <str> subfolder for saving results
CICADA is meant to give way to an interactive design process. Here we show an example of a human user drawing "An avocado chair", and using CICADA to aid her in making the drawing into a "An antique avocado chair". The interface is still under development and will be released shortly.
