Pipeline Core
Python modules and CLI wrappers coordinate stage execution, metadata, and artifacts.
CPU-safe demo for a production-style ML pipeline
Animation LoRA training data, from raw footage to demo-ready artifacts.
This project demonstrates a file-driven computer vision pipeline for preparing animated character LoRA datasets: extraction, tracking, segmentation, pose, embeddings, dataset assembly, inference samples, and animation export.
10/10
66
55
Demo ready
Product results
Public, synthetic outputs that look like a finished ML demo
The generated assets are safe to publish and designed for screenshots, walkthroughs, and portfolio review.
End-to-end flow
Inspectable pipeline stages
The public demo uses synthetic/stub outputs so reviewers can inspect the system without private media, GPU weights, or API keys.
Demo scenarios
Built for screenshots and walkthroughs
These scenarios map the engineering work to what an interviewer can understand in a few minutes.
Deliverables
What this demo now proves
Screenshots and video
Ready-to-share review assets
These files are committed with the static site so the remote repository contains the demo page, screenshots, and a short walkthrough video.
Architecture
Config-driven ML workflow
Model Layer
YOLO, ToonOut/SAM-style segmentation, DWPose, CLIP embeddings, LoRA training, and evaluation adapters.
Data Contracts
YAML/TOML configs and parquet metadata keep stage inputs and outputs reproducible.
Demo Layer
Static JSON powers this public-facing walkthrough, while the real pipeline remains file-based.
Runbook
Local demo commands
pip install -r requirements/core.txt -r requirements/dev.txtbash bash/run_full_pipeline_stub.shpython scripts/demo/run_demo_pipeline.py --skip-pipelinepython -m http.server 8080 -d portfolio-web