• 6 min read
How to Run Jupyter Notebooks on an Apple M1 Mac
Learn how to run Jupyter Notebooks on Apple M1 Macbooks.
• 3 min read
Generating Renaissance Art with Computer Vision
SUMMARY
Two high school students built a DCGAN during a 24-hour hackathon to generate abstract images in the style of Renaissance paintings, training on roughly 300 images sourced from the web. They used Roboflow to preprocess and scale images to 32x32 pixels, and applied data augmentation (reflections and rotations)
• 2 min read
Revamping Train, Validation, Test, Split Management
Splitting data into train, validation, and test splits is essential to building good computer vision models. Today, we are announcing in-app changes to Roboflow that make it even easier to manage your train test splits as you are working through the computer vision workflow.
• 6 min read
Google Researchers Say Underspecification is Ruining Your Model Performance. Here's Five Ways to Fix That.
We read that Google underspecification paper so you don't have to.
• 4 min read
Bringing Street Murals to Life with Computer Vision
SUMMARY
Software engineer Yuri Fukuda built a mobile computer vision app that lets users point a phone at the 3,300-square-foot When Women Pursue Justice mural in Brooklyn and identify the 90 women depicted. She captured photos from multiple angles, labeled them, applied image augmentation in Roboflow to
• 6 min read
YOLOv4 - Ten Tactics to Build a Better Model
The YOLO v4 repository is currently one of the best places to train a custom object detector, and the capabilities of the Darknet repository are vast. In this post, we discuss and implement ten advanced tactics in YOLO v4 so you can build the best object detection model from your custom dataset.
• 3 min read
Hands on with the Roboflow Infer Web Application Interface Builder
SUMMARY
After training a model with Roboflow Train, the Example Web App gives an immediate browser-based interface to test predictions: upload an image or provide a URL, adjust confidence thresholds and bounding box overlap, filter classes, and view results as an annotated image or raw JSON. Because the app
• 5 min read
Occlusion Techniques in Computer Vision
SUMMARY
Computer vision models trained on clean, unobstructed images often fail when objects are partially hidden at inference time. Occlusion-based data augmentation techniques, including Random Erase, Cutout, Hide and Seek, and Grid Mask, address this by blocking portions of training images and forcing models to identify objects from context
• 3 min read
Evaluating Object Detection Models with mAP by Class
SUMMARY
Overall mAP scores can mask weak spots in a multi-class object detection model: a 98% aggregate hides the fact that one class is dragging performance down. Roboflow Train addresses this by surfacing per-class mAP on both the validation and training sets after every training run, so you
• 3 min read
How to Use the Detectron2 Model Zoo (for Object Detection)
SUMMARY
The Detectron2 model zoo provides pre-trained checkpoints for object detection, semantic segmentation, and keypoint detection, all built on PyTorch and maintained by Facebook AI Research. Swapping from the default Faster R-CNN X101 to any other detection model in the zoo requires changing a single config file path
• 5 min read
How This Fulbright Scholar is Using Computer Vision to Protect Endangered Species
SUMMARY
Kasim Rafiq, a Fulbright Scholar and National Geographic Explorer at UC Santa Cruz, is building a vehicle-mounted computer vision system that automatically detects and photographs wildlife during safari game drives, then uses those detections for species population surveys at a fraction of the cost of traditional field methods.
• 2 min read
Train Test Split Guide and Overview
SUMMARY
Splitting a dataset into training, validation, and test sets is a core practice for detecting overfitting and ensuring a computer vision model generalizes beyond its training examples. Roboflow defaults to a 70/20/10 split at upload time but gives you direct control: you can assign individual image batches
• 6 min read
Luxonis OAK-D - Deploy a Custom Object Detection Model with Depth
We are pretty excited about the Luxonis OpenCV AI Kit (OAK-D) device at Roboflow, and we're not alone. Our excitement has naturally led us to create another tutorial on how to train and deploy a custom object detection model leveraging Roboflow and DepthAI, to the edge, with depth, faster.
• 6 min read
Behind the Design of an Augmented Reality Board Game App
During the summer of 2019, I received a Facebook message from Roboflow co-founder Brad Dwyer asking me if I wanted to design a new mobile app he was working on. His previous app, Magic Sudoku, had recently generated some buzz online and even won a Golden Kitty award from
• 5 min read
How to Save and Load Model Weights in Google Colab
SUMMARY
Google Colab disconnects sessions after 12 hours or an hour of browser inactivity, which means a training run can complete but its weights are lost if you do not export them before the session ends. This post explains how to identify the weights file path after training, download it
• 9 min read
Using Computer Vision to Help Deaf and Hard of Hearing Communities
SUMMARY
Data scientist David Lee built a computer vision model to recognize American Sign Language alphabet from images, using data warping and oversampling to handle the limited and imbalanced training data. The goal is to give deaf and hard-of-hearing communities a tool that can interpret ASL through a
• 7 min read
An Introduction to the COCO Dataset
SUMMARY
The Microsoft Common Objects in Context (COCO) dataset is the standard benchmark for evaluating computer vision models, containing over 330,000 images annotated across 80 object and 91 stuff categories. It supports object detection, semantic segmentation, and keypoint detection tasks, and its scale and contextual diversity make it more
• 31 min read
Software Engineering Daily Podcast
Roboflow co-founder Brad Dwyer was a guest on the Software Engineering Daily podcast. Listen on your favorite podcast app (Apple Podcasts, Spotify, Overcast, Stitcher), or see the full transcript below.
Software Engineering Daily: Roboflow: Computer Vision Models with Brad Dwyer on Apple PodcastsTraining a computer vision model is not
• 2 min read
Introducing an Improved Shear Augmentation
Today, we introduce a new and improved shear augmentation. We'll walk through some details on the change, as well as some intuition and results backing up our reasoning.
• 2 min read
Fighting Wildfires with Computer Vision
SUMMARY
Abhishek Ghosh, a graduate student at Texas A&M working with AI For Mankind, trained an EfficientDet D0 computer vision model to detect early smoke from forest fires using images from weather towers. Using Roboflow Pro, he applied advanced augmentations and a static crop along the horizon line
• 6 min read
How Tesla Teaches Cars to Stop
SUMMARY
Tesla Senior Director of AI Andrej Karpathy's CVPR 2020 talk reveals that even a seemingly straightforward computer vision task like stop sign detection involves dozens of edge cases: occluded signs, non-standard mounting, signs that do not require stopping, and conditionally active signs. Tesla handles this by
• 1 min read
Roboflow Wins $15,000 Award
Roboflow is honored to be named as one of Iowa's most promising startups and one of the top 3 winners of the Pappajohn Entrepreneurial Venture Competition. The annual awards honor companies with innovative technology that have grown quickly.
Since launching in January, Roboflow has helped over 8,000
• 24 min read
Glossary of Common Computer Vision Terms
Computer Vision (and Machine Learning in general) is one of those fields that can seem hard to approach because there are so many industry-specific words (or common words used in novel ways) that it can feel a bit like you're trying to learn a new language when you're trying to get started.
• 3 min read
Improving Cancer Research with Computer Vision
SUMMARY
Researcher Mateo Sokac at Aarhus University is using object detection to identify neutrophils in dry microscopy images, with the goal of accelerating cancer research in humans. Starting from 127 labeled images, augmentation via Roboflow expanded the dataset to 451 samples and lifted mean average precision to 0.70, a
Stay Connected