Blog

Kelly M.

Kelly has a background in both software engineering and data science and is a Developer Advocate at Roboflow.

Latest Posts by Kelly M.

Launch: Instance Segmentation Project Training and Inference

19 May 2022 • 2 min read

Launch: Instance Segmentation Project Training and Inference

SUMMARY Roboflow now supports instance segmentation project training and inference natively on the platform, letting teams train directly from a pre-trained COCO checkpoint without writing training code. Instance segmentation produces per-object outlines rather than just bounding boxes, making it the right choice when downstream tasks require measuring object

How to Train a YOLOv5 Oriented Bounding Box Model

11 May 2022 • 3 min read

How to Train a YOLOv5 Oriented Bounding Box Model

Learn how to train a YOLOv5 Oriented Bounding Box (OBB) object detection model.

cinco de mayo multi-label classification

5 May 2022 • 3 min read

Cinco de Mayo, Beer and Taco Dataset with Multi-Label Classification

SUMMARY Multi-label classification lets a single model predict multiple classes per image simultaneously, making it a natural fit for scenes where counting or locating objects precisely is less important than knowing what is present. This post builds a taco and beer classifier by merging several Roboflow Universe datasets (including

Building a U.S. License Plate Detection Model And Sharing It On Roboflow Universe

8 Mar 2022 • 5 min read

Building a U.S. License Plate Detection Model And Sharing It On Roboflow Universe

SUMMARY This post walks through building a U.S. license plate detection model trained with Roboflow, achieving 94.9% mAP, 95.7% precision, and 90.4% recall on images collected from Google and Central Florida locations. The workflow covers annotation, preprocessing with auto-orient and resize, augmentations tuned for occlusion

How To Avoid Bias In Computer Vision Models

14 Feb 2022 • 5 min read

How To Avoid Bias In Computer Vision Models

SUMMARY Bias in a computer vision model is the average error that persists after training, and it almost always traces back to data quality rather than model architecture. The practical mitigation steps covered here include collecting large and representative datasets, using active learning to surface gaps, running model error analysis,

How to Train Computer Vision Models on Aerial Imagery

7 Feb 2022 • 7 min read

How to Train Computer Vision Models on Aerial Imagery

SUMMARY Aerial imagery from drones and planes presents distinct challenges for computer vision: small objects have few pixels to form features from, large image sizes slow training and inference, and orientation bias can mislead models. This post walks through the full pipeline for building a custom model on aerial data

Stay Connected

Get the Latest in Computer Vision First