computervision-recipes

Why we included this project

Teams building custom vision models in PyTorch will find this more useful than a bare code dump. The notebooks carry you through a complete project, starting with fine-tuning a classifier on your own images and then branching into similarity search, object detection, keypoints, segmentation, and video tasks such as action recognition and tracking. Instead of re-implementing algorithms, the repo layers utility code over proven libraries for loading image data, tuning, evaluation, and scaling to the cloud, so you can get from a business problem to a deployed solution faster. The examples also serve as reference material on common pitfalls and best practices, which helps when your team is newer to vision workflows. If Azure is part of your stack, the operationalization guides cover deployment and when a prebuilt API saves you from training a custom model at all.