How do you cluster images at scale?
mediumAnswer
- (1) Feature extraction: pass through pretrained CNN or ViT (DINOv2, CLIP) → embeddings.
- (2) UMAP / PCA to ~50 dim for compute.
- (3) HDBSCAN or mini-batch k-means.
- (4) For huge scale (billions): FAISS + BIRCH or streaming k-means.
- (5) Semi-supervised: use CLIP embeddings + text queries as prompts for prototype clustering.
- Modern: DINOv2 embeddings + k-means or HDBSCAN is production default.
Check yourself — multiple choice
- Same as k-means only
- Pretrained CNN/ViT embeddings (DINOv2, CLIP) → UMAP/PCA → HDBSCAN / mini-batch k-means; FAISS + BIRCH at billions; DINOv2 + k-means production default
- Random
- Not scalable
Image clustering: DINOv2/CLIP embeddings + UMAP + HDBSCAN/k-means.
#applications#clustering
Practise Unsupervised Learning
214 interview questions in this topic.