Member-only story

How Pictures Structural Similarity can boost your Computer Vision projects

Discover why this skimage function is more powerful than MSE and how it can help you accelerate pictures’ classification and avoid target leakage when training your deep learning network.

Pierre-Louis Bescond
Towards Data Science
6 min readFeb 1, 2021

--

Photo by Ross Joyner on Unsplash

If you have been working on image recognition or computer vision projects before, you know that having a properly classified dataset is of the utmost importance!

And I am not talking here about “ready-to-use datasets” such as CIFAR or MNIST but real and raw pictures that you have to sort manually before conducting any training.

I was working on a new computer vision initiative lately and discovered how the “Structural Similarity” (SSIM) between pictures can help boost the sorting process and also avoid some common pitfalls.

Let’s dive in!

You should always question great accuracy

I know this is unusual but I will start by the end… I was done with a deep learning network training and quite proud of the accuracy it had achieved, close to 100%.

Sometimes Life gives you lemons… but sometimes you also receive really clean and distinct…

--

--

Towards Data Science
Towards Data Science

Published in Towards Data Science

Your home for data science and AI. The world’s leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals.

Pierre-Louis Bescond
Pierre-Louis Bescond

Written by Pierre-Louis Bescond

Head of Data & Advanced Analytics @ Roquette | Winner of the 1st WorldWide Data Centric Deep Learning Contest | Data Science & Machine Learning Passionate!

Responses (1)