Latest Posts

27 Oct 2022 • 4 min read

Blurring Faces to Preserve Privacy with Computer Vision

SUMMARY Automatically blurring faces and license plates before data enters a storage pipeline is a practical way to handle personally identifiable information in computer vision deployments, following the same approach used by Google Maps Street View. This tutorial covers collecting face and license plate datasets from Roboflow Universe, training a

26 Oct 2022 • 4 min read

Launch: Semantic Segmentation for Labeling, Training, Deployment

SUMMARY Semantic segmentation assigns a class label to every pixel in an image, making it useful for tasks where per-pixel precision matters: road and sidewalk mapping from aerial imagery, crack detection in infrastructure, defect identification in manufacturing, and medical image analysis. Roboflow now supports semantic segmentation end-to-end,

21 Oct 2022 • 8 min read

What is an Autoencoder?

SUMMARY An autoencoder is a neural network that compresses an input into a compact latent vector and then reconstructs it, with the latent representation, not the output, being the useful artifact. Six main variants exist for computer vision: undercomplete, denoising, sparse, contrastive, Variational (VAE), and Vector Quantised-Variational (VQ-VAE)

21 Oct 2022 • 6 min read

What is Object Tracking in Computer Vision?

Learn about object tracking and how it is used in computer vision.

20 Oct 2022 • 5 min read

Top 6 Manufacturing Datasets for Computer Vision

Explore six manufacturing datasets you can use for computer vision projects.

20 Oct 2022 • 7 min read

Accelerate PyTorch Models via OpenVINO™ Integration with Torch-ORT

SUMMARY OpenVINO integration with Torch-ORT lets PyTorch developers run accelerated inference on Intel CPUs, integrated GPUs, and VPUs without changing their existing code structure, requiring only two additional lines of Python. This post benchmarks a custom-trained YOLOv7 model on a dataset from Roboflow Universe and shows roughly a

18 Oct 2022 • 8 min read

How to Use Computer Vision to Control OBS Studio

Learn how to use computer vision to control OBS Studio.

14 Oct 2022 • 4 min read

Launch: Cloning Images from Open Source Datasets

SUMMARY Roboflow Universe now lets you clone any of its 100+ million open source images directly into your own Roboflow project, with or without their existing annotations. This removes one of the biggest time sinks in computer vision development: sourcing labeled training data from scratch. You can filter by object

14 Oct 2022 • 2 min read

Roboflow Joins AWS Global Startup Program and Becomes AWS ISV Accelerate Partner

SUMMARY Roboflow has joined the AWS Global Startup Program and become an AWS ISV Accelerate Partner, deepening its integration with the cloud infrastructure that already powers Roboflow's model pipeline. The ISV Accelerate designation puts Roboflow alongside companies like Databricks and Auth0 in a co-sell program for enterprise

13 Oct 2022 • 4 min read

Sharing Your Computer Vision Project on Roboflow Universe

In this guide, learn how to launch your computer vision project on Roboflow Universe.

13 Oct 2022 • 3 min read

Launch: Use YouTube Videos for Training Data and Inference Testing

SUMMARY Roboflow now accepts YouTube video URLs directly as an upload source, letting you sample frames from any of YouTube's 800+ million videos to build or expand a computer vision training dataset without downloading files manually. The same feature works in reverse: you can paste a YouTube URL

1 Oct 2022 • 4 min read

Roboflow Changelog: September 2022

Each month, we share a recap of product and company updates. In this post, we detail Roboflow's updates from September 2022.

28 Sep 2022 • 4 min read

Launch: Smart Polygon Labeling

SUMMARY Roboflow Annotate's Smart Polygon tool uses a machine learning model behind the scenes to generate a polygon label with a single click on an object, reducing what previously required many manual point placements to a near-instant operation. Precise polygon annotations are required for training instance segmentation

28 Sep 2022 • 6 min read

Object Detection vs. Image Classification vs. Keypoint Detection

Learn what object detection, classification, and keypoint detection are, and when tech technique is useful.

28 Sep 2022 • 9 min read

Supervised Learning vs. Unsupervised Learning: Explained

Learn what supervised and unsupervised learning are and how they compare.

21 Sep 2022 • 12 min read

How to Measure Distance in Photos and Videos Using Computer Vision

Learn how to measure distance in photos and videos using computer vision.

21 Sep 2022 • 10 min read

How to Train YOLOv5 Instance Segmentation on a Custom Dataset

Learn how to train a YOLOv5 instance segmentation model on a custom dataset.

20 Sep 2022 • 2 min read

Launch: Test Computer Vision Models Locally

SUMMARY Roboflow Inference can now run locally on a development machine via a single npx command, with no CUDA installation or Docker setup required. Any 64-bit Intel, AMD, or Arm CPU (including M1 Macs and Raspberry Pi 4) can host the inference server on port 9001, and any pre-

14 Sep 2022 • 9 min read

Using Computer Vision with Drones for Georeferencing

SUMMARY A computer vision model can detect objects in drone footage but only tells you where something appears in a frame, not where it is on Earth. This tutorial shows how to combine detection output with the flight log from a DJI Mavic Air 2, converting pixel coordinates to latitude

9 Sep 2022 • 4 min read

Roboflow Changelog: July & August 2022

Each month, we share a recap of product and company updates. In this post, we detail Roboflow's updates from July and August 2022.

7 Sep 2022 • 2 min read

Roboflow now available in the AWS Marketplace

SUMMARY Roboflow is now available through the AWS Marketplace, letting AWS customers purchase the platform under consolidated billing and apply spend toward AWS Enterprise Development Program commitments. Available tiers include Roboflow Growth (25k source images, 100k generated images, 1,000 monthly API calls) and custom enterprise orders with added usage

7 Sep 2022 • 7 min read

How I use Computer Vision and Twilio to Guarantee Availability at Busy Public Tennis Courts

SUMMARY A non-ML practitioner built a working computer vision application that monitors a local tennis court via webcam, detects whether players are on the court using a model trained in Roboflow, and sends an SMS notification through Twilio when a court opens up. The post covers the full pipeline:

5 Sep 2022 • 5 min read

Train Activity Recognition Models Using Spectrograms and Computer Vision

As Ph.D. students in the Active Robotics Sensing Lab (ARoS) at NC State under the supervision of Dr. Edgar Lobaton, we developed in conjunction with The Engineering Place at NC State a set of activities to walk high school students through the entire computer vision pipeline.

30 Aug 2022 • 7 min read

WTF COCO - The Weird Images that Underpin Modern Computer Vision Models

COCO is an industry standard dataset for benchmarking the performance of object detection models. The dataset was created by "gathering images of complex everyday scenes containing common objects in their natural context" and contains image annotations in 91 categories, with over 1.5 million object instances. People have