Datasets:
Use PE-Core-L14-336 embeddings and add caption similarity plugin
#2
by JonasWurst - opened
- Register a larger Perception Encoder (PE-Core-L14-336) for video embedding and text search; the default PE-Core-T16-384 separates captions poorly on these short clips.
- Add a GUI plugin that stores video-caption cosine similarity as video metadata, to find clips whose narration doesn't fit.
- Fix scripts for the current LightlyStudio API (
dataset.collection_id).
🤖 Generated with Claude Code
igor-lightly changed pull request status to merged