←
Ultralytics4,745 posts
Ultralytics avatar

Ultralytics

@ultralytics

Simpler. Smarter. Further.

◷ Joined February 2014▧ 1,978 Photos and videos
Ultralytics avatar
Ultralytics@ultralytics·40m

Image rotation is live on Ultralytics Platform 🔄 Open a sideways or upside-down image and, with edit access, press R: the image and its labels turn together. The original stays stored, so four turns restore it exactly, and uploads tagged with EXIF orientation now arrive upright. Learn more ➡️ docs.ultralytics.com/platfor…

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·2h

Running Ultralytics YOLO26n on a Raspberry Pi 5 just got 84% faster! 🚀 Arm Software Developers reports reaching around 27 inference FPS on the CPU alone using their new open-source runtime. See the results ➡️

Real-time object detection, running entirely on the CPU. ⚡️ Arm's new open-source runtime delivers around 27 inference FPS for @Ultralytics YOLO26n on @Raspberry_Pi 5, with no accelerator required. That's roughly 84% higher FPS than the fastest general-purpose runtime in Arm's benchmarks, unlocking more performance and CPU capacity for edge AI applications. okt.to/2kSrzf

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·22h

One camera. Multiple conveyors. Separate counts for every line. 📦 With Ultralytics YOLO26, packages can be counted as they enter defined conveyor zones. This is very useful for monitoring flow, throughput, and line activity. Read more ➡️ docs.ultralytics.com/guides/…

0:25
Ultralytics avatar
Ultralytics@ultralytics·23h

Segment 1,000 objects per image 🎯 Ultralytics 8.4.213 decodes large mask stacks inside each box. In the PR's YOLO26n-seg benchmark (synthetic tiled 1080p scene, imgsz=1280, retina_masks=True, RTX PRO 6000), 1,000 masks went from 821 ms to 56 ms, prediction plus summary. Details ➡️ github.com/ultralytics/ultra…

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·Nov 17

Search for a model by what it detects, not just by its name 🔎 The new Models tab in Ultralytics Platform Explore finds public trained models for "forklift" or "hard hat" from what their training images show. Test one in Predict, then clone it to fine-tune or deploy. Explore ➡️ docs.ultralytics.com/platfor…

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·Nov 17

Code 👇 """""" from ultralytics import solutions searcher = solutions.VisualAISearch( data="images", # replace with a path to build the search engine with your own images device="cpu", # configure the device for processing, e.g., "cpu" or "cuda" ) results = searcher("a dog sitting on a bench") # Ranked Results: # - 000000546829.jpg | Similarity: 0.3269 # - 000000549220.jpg | Similarity: 0.2899 # - 000000517069.jpg | Similarity: 0.2761 # - 000000029393.jpg | Similarity: 0.2742 # - 000000534270.jpg | Similarity: 0.2680 """"""

Ultralytics avatar
Ultralytics@ultralytics·Nov 16

Read text from photos, scans and PDFs in a few lines of Python 📝 Our new guide runs Ultralytics OCR with YOLO26x-Text and PP-OCRv6: words come back in reading order at any rotation, and a TensorRT engine reads dense photos 2.5x to 3x faster. Try it ➡️ ultralytics.com/blog/yolo-oc…

Post attachment
Muhammad Rizwan Munawar avatar
⟳ Ultralytics retweeted
Muhammad Rizwan Munawar@muhammdrizwanmr·Nov 14

Decide which potatoes pass quality checks! 🥔 With @ultralytics YOLO26, potatoes moving on a conveyor belt can be detected and classified as good or bad in real time. Instead of manual sorting, visual inspection happens automatically as products flow through the line. This enables faster quality control, consistent grading, and reduced waste in food processing facilities. #PotatoPlease #QualityControl #AI

0:12
Ultralytics avatar
Ultralytics@ultralytics·Nov 14

Code 👇 """"" from ultralytics.data.utils import visualize_image_annotations label_map = { # Define the label map with all annotated class labels. 0: "person", 1: "car", } # Visualize visualize_image_annotations( "path/to/image.jpg", # Input image path. "path/to/annotations.txt", # Annotation file path for the image. label_map, ) """""

Ultralytics avatar
Ultralytics@ultralytics·Nov 13

Your originals stay in your S3 bucket. Your model ends up behind an HTTPS endpoint 📦 Our new guide connects an Amazon S3 bucket to Ultralytics Platform, trains YOLO26 without a manual upload and deploys a dedicated endpoint in one of 42 regions. Follow the steps ➡️ ultralytics.com/blog/train-y…

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·Nov 13

Bad annotations can lead to poor model predictions. 🔍 With Ultralytics, use the "visualize_image_annotations" method to inspect bounding boxes and class labels before training. Catch labeling errors early and improve dataset quality. Read more ➡️ docs.ultralytics.com/usage/s… Curious how it works? Check out the Python code snippet below! 👇

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·Nov 13

Faster YOLO26 inference on Axelera Metis! 🚀 With Voyager SDK 1.9, all 7 Ultralytics YOLO26 tasks now support export, prediction, and validation. ⚡ COCO128 benchmarks show 109 images/s at batch 32, up from 40 on SDK 1.8 — a 2.7× throughput boost! 🔗 github.com/ultralytics/ultra…

Post attachment
Ultralytics avatar
Ultralytics@ultralytics·Nov 13

Sideways scan? Upside-down page? Our OCR demo reads it upright 📄 Ultralytics 8.4.210 pairs YOLO26x-Text with PP-OCRv6 and returns text in reading order, about 1 s per page on a sample PDF on GPU. Try it ➡️ ultralytics.com/use-cases/do…