MSc dissertation

image-based music genre classification · CNNs + Concept Whitening

home Inkling Kimi K3 compare study path sources MSc

---

Classify music genres from vinyl album covers with CNNs. Two models: batch-norm baseline and Concept Whitening for interpretability. CW improves feature transparency without a large accuracy loss.

Where it lives

• Study monorepo: dissertation/main.ipynb in xai-studies and private lyffseba/xai
• Public archive: msc-dissertation

Run (not offline-stdlib)

1. Open main.ipynb
2. Point dataset paths at your local cover images
3. Run cells (TensorFlow / Keras required for execution)

Related experiment (not merged into xai): optional Mojo/MAX port lives separately on disk.