← Back [EVENT] October 2025 · Conference Talk

Codemotion 2025

From pixels to features: one backbone to rule them all! A journey through visual feature extraction — from hand-crafted descriptors to foundation models.

Conference Talk · Codemotion · October 28–29 · Milan

From Pixels
to Features

One backbone to rule them all.

A journey through visual feature extraction

from hand-crafted descriptors to foundation models.

— Codemotion · October 28–29, 2025 · Milan

Photo Gallery

At the Conference

Speaking at Codemotion 2025
Engaging with the audience
Presenting slides
Full stage view
Q&A session
About the Talk

Talk Summary

CORE THEME

We explored what a backbone is, how it turns pixels into a reusable feature space, and how to harness it to prototype quickly.

Essential theory (how they're built, why they work) × hands-on practice (how to use them).
Minimal code patterns — from feature extraction to small pilot tasks, showing how the same representations can power multiple use cases.

From SIFT to Foundation Models

The talk traced the evolution of visual features: from SIFT and HOG to deep CNNs, then to Vision Transformers and today's foundation models (CLIP, DINOv2, SAM). Each step redefined what "understanding an image" means.

Key takeaway: With modern backbones, you no longer need millions of labeled samples to solve a new vision task — you just need the right representation.

← Back to Talks