Detecting Anomalous Images with CLIP and Isolation Forest
July 15, 2026Finding suspicious images in a dataset using an unsupervised anomaly detection pipeline with CLIP embeddings, multi-dimensional PCA sweeps, and per-class Isolation Forests.
A digital notebook and portfolio documenting hands-on experiments, research, and projects in machine learning and software engineering.
Recent insights and articles
Finding suspicious images in a dataset using an unsupervised anomaly detection pipeline with CLIP embeddings, multi-dimensional PCA sweeps, and per-class Isolation Forests.
Building a robust classification pipeline using clustering centroids to categorize unseen images, filter out anomalies with confidence thresholds, and benchmark against zero-shot CLIP.
Combining modern embeddings with traditional ML: how to use CLIP, PCA, and K-Means for unsupervised image labeling.
How I Merged Multiple Git Repositories Without Losing History. A Guide to Consolidating My Android Projects
In this article, we will explore how indexing works in SQLite, differentiate the performance when applying indexing, and experiment with scenarios where it may break the utilization of the index.
Explore all articles and tutorials