Ultralytics YOLOv8.2 ReleaseUltralytics YOLOv8.2 ReleaseUltralytics YOLOv8.2 Release Arrow
Green check
Link copied to clipboard

Ultralytics YOLOv8 Turns One: A Year of Breakthroughs and Innovations

On the 1st anniversary of Ultralytics YOLOv8 we reflect on its impact, where to find all the documentation, how train models and so much more!

Today, January 10th, 2024, marks a year since the launch of Ultralytics YOLOv8, and it’s time to celebrate! It’s been an exciting year of milestones and pushing the boundaries of what’s possible. Join us as we revisit the highlights of 2023 and what’s next in 2024.

Reflecting on YOLOv8's Impact in 2023

Data from the first year of YOLOv8
Fig 1. 19 Million YOLOv8 Models Trained in 2023

YOLOv8 has been welcomed warmly by avid computer vision enthusiasts and the community at large. In the past year, the Ultralytics package has been downloaded more than 20 million times, with a record-breaking 4 million downloads just in December alone. Glenn Jocher, our Founder & CEO, is happy to share that the interest in YOLOv8 continues to grow, with over 1,000 inference jobs initiated every second of every day! 

Beyond evoking intrigue and curiosity, YOLOv8 has also proven itself to be impactful in practical, real-world applications. This year, we have seen 5 million users and 15 billion events benefit from YOLOv8 across various industries and domains. From improving surveillance systems to innovative advancements in healthcare, agriculture or manufacturing, YOLOv8 is revolutionizing industries globally.

Expanding YOLOv8’s Docs

We’re bringing YOLOv8 closer to you! Our documentation is now available in 11 languages, with 200+ docs pages, and is continuously expanding to serve our diverse community's needs better! Our documentation goes above and beyond and consists of guides for the following real-world projects:

The docs also illustrate the support that Ultralytics provides for various datasets. For example, recently, the Open Images V7 dataset with 600 classes was added to the list of supported datasets. Additionally, we have made available a pre-trained model for the Open Images V7 dataset for you to try out!

Creating Custom Trained YOLOv8 Models

Beyond using pre-trained models, users are also looking for custom computer vision solutions that solve very specific business problems. The capability to train YOLOv8 models on custom data has emerged as a significant advantage, and a staggering count of 19 million YOLOv8 models were trained in 2023. These models have been trained for various tasks, with 64% dedicated to object detection, 20% for image segmentation, 15% for pose estimation, and 1% for image classification

In part, these numbers are possible because anyone can train YOLOv8 thanks to Ultralytics’ no-code ML platform, Ultralytics HUB - regardless of their coding expertise. You can quickly create and train advanced models without needing any code on Ultralytics HUB, which is accessible on both the web and mobile. As we celebrate YOLOv8's successes, let’s also look back at how Ultralytics HUB has evolved over this past year.

Ultralytics HUB’s Growth in 2023

2023 has been a great year for Ultralytics HUB, with 84 impactful version updates, each one steering us towards better functionality and user experience. We unveiled major features like ‘Teams’ for seamless collaboration, our Pro HUB version for enhanced capabilities, a clearer billing history for your financial peace of mind, and a new user feedback system.

Managing your models has never been easier. You can now compare and move models when working on a project. We’ve enabled even more formats to offer you flexible model export options, and so much more.

Other than new and improved features, a lot of time and energy has also gone into improving existing features. For instance, thanks to priority loading, the platform starts up as fast as lightning. HUB’s branding and UX have been reimagined for a visually stunning experience, and the user dashboard has quick links and an intro video for a smoother start. 

The API key management has been revamped to be even safer, and the platform has been integrated with the Ultralytics App for a smoother back-and-forth experience. And that's just to mention a few!

Ultralytics HUB screen for users
Fig 2. Enhanced User Dashboard: Quick links and an intro video for a smoother start.

As we step into 2024, we're excited to see HUB grow even further with your continued support and input. Let’s explore what’s next for YOLOv8 together! 

What’s Next for YOLOv8 in 2024

Discover the latest developments, including the YOLOv8.1 release, and explore what's in store for Ultralytics in 2024!

Introducing: Ultralytics Explorer

Just in time for the first anniversary of YOLOv8, we are coming out with a new YOLOv8-enabled tool called the Ultralytics Explorer. This innovative tool promises to change how users explore and interact with their datasets. You can use either the Ultralytics Explorer API or the GUI to be able to filter and search your datasets using SQL queries, vector similarity search, and semantic search. 

One exciting feature of Ultralytics Explorer is image matching. For example, you can select an image in your dataset and find all the images in your dataset that are similar to this image. This can make understanding and managing your dataset easier. 

Fig 3. New Ultralytics Explorer

Let’s say you want to see all the pictures of giraffes in your dataset, you can do it with a few clicks! It also supports multiple image matching, meaning when you select multiple images to match - the average of the images is calculated.

You can also write SQL queries to find a specific number of images in the dataset with specific labels. This can come in handy when you want to see a sample of 10 images from the dataset with a label like ‘dog’. It helps you get an idea of the data that has been annotated. 

Another exciting feature is the Ask AI feature. In case, you aren’t proficient in SQL, It allows you to use the querying feature without the need for SQL. For example, you can ask our AI-powered query generator to show you 100 images with exactly one person and 2 dogs, and it'll internally generate the query and show you the query results. 

Ayush Chaurasia, Advisor at Ultralytics, said, ‘The best part is that because the Ultralytics Explorer API itself is open source, you can use the API to create applications for dataset validation, exploration, and more.’ Check out more details about the Ultralytics Explorer here.

A New YOLOv8 Task: Oriented Object Detection

YOLOv8 takes a significant leap forward by introducing Oriented Object Detection, also known as OBB. This advanced feature is designed to deliver precise detection results, particularly for objects at various angles and rotations. 

This improves the robustness and reliability of detection, especially for inclined objects like aerial remote-sensing images and text detection. OBB stands out by its ability to accurately locate objects in images, minimizing the inclusion of background areas. This precision significantly enhances object classification by reducing background noise.

Aerial scene of yachts with Oriented Bounding Boxes (OBB)
Fig 4. An example of an aerial scene with Oriented Bounding Boxes (OBB).

Jing Qiu, ML Engineer at Ultralytics, shares insights on our latest innovation: 'At the heart of the new YOLOv8-OBB model lies the robust foundation of our YOLOv8 detection model. While it incorporates additional parameters and computation, we’ve ensured that its inference speed remains swift for real-time applications, mirroring the performance of our standard detection models. It is user-friendly and shares the same API but is marked by a simple 'obb' sign, making it extremely easy to train, validate, predict, and export, similar to our other tasks.'

We’re also excited to announce added compatibility for training a model on the DOTA v2 dataset. Dive into more details here and explore how this expands the capabilities of YOLOv8.

Improved Image Classification Models

While adding new tasks for YOLOv8 to support is essential, it is equally vital to improve and enhance the original tasks. Echoing this sentiment, the image classification task supported by YOLOv8 has been improved.

Fatih Akyon, ML Engineer at Ultralytics, highlights, 'We have integrated SOTA classification augmentations into the Ultralytics training pipelines. This helps improve the classification scores. The base yolov8-classification models were retrained with the new pipeline.”

Performing image classification
Fig 5. An image showcasing image classification.

To learn more about YOLOv8’s ability to classify images, check out this docs page. 

We’re Accepting User Contributed Examples!

One of the main successes of YOLOv8 in 2023 has been the amount of love, support, and contributions from our community. With over 225 contributions so far, we’re thankful for each one that has helped refine and enhance YOLOv8. Your valuable input has driven us to refine and fine-tune YOLOv8, making it more adaptable and responsive to varying needs and challenges across diverse industries. 

As we enter 2024, we are excited to expand our repository of user-contributed examples. Your contributions are pivotal in addressing real-world scenarios where computer vision can be a solution. We invite you to collaborate by sharing your innovative use cases, success stories, and unique implementations with the wider YOLOv8 community. Your contributions inspire fellow enthusiasts and guide YOLOv8 toward new heights.

Together, let's build a vibrant repository of user-contributed examples that showcase YOLOv8's versatility and reflect the creativity of our community. You can find more examples and contribute to our repository here. If you have any questions about contributing, our guide is there to help you.

Thank you for your unwavering support, and we look forward to witnessing the incredible year ahead for YOLOv8. Stay tuned for more updates, innovations, and collaborative achievements. Here's to an amazing year ahead! 🚀

Facebook logoTwitter logoLinkedIn logoCopy-link symbol

Read more in this category

Let’s build the future
of AI together!

Begin your journey with the future of machine learning