Do you have data challenges like multispectral images, noisy labels, or small training sets?
Are your machine learning models robust and unbiased? Are you confident that they will generalize to images from different sources?
Sometimes data challenges are easy to decipher and handle, but other times they are more subtle.
There could be an obvious distribution change causing degraded performance. But it can also be hidden or caused by multiple factors.
Gathering a large, diverse training set may be the ideal solution but is often not feasible.
What do you do?
Perhaps you have a few techniques in your toolkit like normalization to handle the obvious changes in image color or augmentation to create extra synthetic training examples.
But have you ever taken a step back to fully understand the challenges in your data and to be sure that you’re tackling each with data- and modeling-centric techniques?