[MM'22 Oral] AI Illustrator: Translating Raw Descriptions into Images by Prompt-based Cross-Modal Generation

This reposity is the official implementation of AI Illustrator: Translating Raw Descriptions into Images by Prompt-based Cross-Modal Generation.

The proposed pipeline is shown below.

Currently, we support pretrained models on 3 domains: Human face, Cat and Church. The download urls are:

The default path of pretrained models is ./pretrained_projectors.

After downloading the pretrained models, you can simply generate images by command

 python single_generate.py --kind <domain> --projector_path <path/to/the/pretrained_projector> --save_path <path/to/the/save_dir> --strength 1.75 --prompt_path <path/to/the/text_prompt>

One example is

 python single_generate.py --kind 'human' --projector_path './pretrained_projector/c2s_human.pth' --save_path './outputs' --strength 1.75 --prompt_path './prompts/ffhq_text_prompt.pth'

The values of argument "kind", "projector_path" and "prompt_path" should match. By default, "kind" should be one of "human", "cat" and "church".

100 raw descriptions can be found in ./benchmark/description_benchmark_100.txt

Corresponding generation results of our method can be download at