How I found my Pokémon cards thanks to Postgres: an AI journey
summary
A complete pipeline identifies and finds specific Pokémon cards by combining object detection, embedding generation and vector search: images are labeled with Label Studio, a ResNet50 fine-tune from Hugging Face detects card bounding boxes, CLIP produces 768‑dim CLIP embeddings, and embeddings are stored and queried in PostgreSQL using pgvector. The demo runs on a webcam, exact nearest‑neighbor scans give the most accurate matches (at cost of latency), and IVF/NHNSW indexes are available when speed matters.
AI-written summary · about
chapters (8)
- 0:00Introduction and motivation
- 1:34Defining the problem
- 3:26Identifying a solution
- 8:43Building the AI model
- 16:51Generating and storing vectors
- 25:55Similarity search in Postgres
- 35:55Demo and real-world results
- 37:03Future work and Q&A
