Obsidian Source: Notes / Sakana - Task 2 (Generative AI)
Summary
Pending synthesis from local Obsidian source.
Original source title: Sakana Task 2 (Generative Ai)
Extracted Preview
We'll discuss in detail what we need for this project.
*Problem : Train/Fine-Tune a Stable Diffusion model to produce novel Kanji from English definition.*
We WANT the model to "hallucinate" new Kanji from novel English definitions. We have to build a dataset first. Stroke number should not be printed, and
Tasks : Given are the data from the links(SVG files). We want to render the fonts in pixel format. We should be able to build a dataset of English meaning from kanjidic.xml for each Kanji character and map the image to the meaning for training/fine-tuning stable diffusion pipeline.
Using our fine data engineering skills will come in handy.
Integration Notes
- Source folder:
/home/yashs/Documents/Docs/Obsidian/Research-Notes - Local source:
/home/yashs/Documents/Docs/Obsidian/Research-Notes/Notes/Sakana - Task 2 (Generative AI).md - Raw copy:
raw/obsidian/research-notes/Notes/Sakana - Task 2 (Generative AI).md