For many companies, the holidays are a lull and they end the year with a whimper. At Roboflow, shipping is a huge part of our culture, so we wanted to start the year off with a bang.
We coined the 12 Days of #Shipmas and shipped 12 new features in 12 days. New updates were released to improve model-assisted labeling, model training, Roboflow Universe, our annotation tools, the REST API, Python SDK, and more.
Day 1: Remapping in Label Assist
Label Assist is the feature in Roboflow Annotate that enables you to use a computer vision model to pre-annotate your images so you can spend your annotation time correcting a model vs labeling from scratch.
But if your model is trained on different classes than you're labeling it isn't as useful as it could be... until now! With this new feature you can remap a model's classes to your dataset's ontology. This is especially useful if you're using a model from one dataset to bootstrap another.
Day 2: Starring Projects on Roboflow Universe
Roboflow Universe is our community for computer vision projects. It hosts over 100 million images, 100,000 datasets and 10,000 fine-tuned models. You can now show your support for projects & bookmark them for later by starring them (just like you can star open source projects on GitHub).
Day 3: Training from Public Roboflow Universe Checkpoints
Those 10,000+ fine-tuned models on Universe just became a lot more useful for bootstrapping your own computer vision models. You can now use them as a transfer learning checkpoint. This means your model starts with existing knowledge it's learned from a prior training run.
If someone has trained and shared a model on a similar domain as your project, you can star it on Universe and then use its weights to kickstart your own model's training.
Day 4: Over 10,000 Models for Label Assist
You can now also use those 10,000+ Universe models with Label Assist to annotate additional images for your dataset. This, combined with class remapping in Label Assist, supercharge Roboflow Annotate to help you jumpstart a new computer vision project in record time.
Day 5: Polygon Annotation Updates
Our polygon tool got some major updates including better support for large aerial images, faster performance, several bug fixes, and new UI affordances to make it easier to use.
Day 6: Roboflow Universe Dataset Cross-Linking
It's now easier to share your work with your friends and followers with cross-linking between the editable of your project and the public view on Roboflow Universe. Show off your hard work by sharing your datasets & models.
Creating and sharing your project on Roboflow Universe can be a great way to build your brand and boost your resume as a computer vision expert. We've even seen community members get hired or promoted based on the cool projects they've built!
Day 7: Read/Write API Methods for Projects
We're making it easier to build automated MLOps pipelines with new API methods (and corresponding
roboflow Python package features) for creating versions, exporting datasets, and training models. Check out the full details in our Code Actions launch post.
Day 8: Label Datasets via API
Continuing on with the API automations theme, we've expanded our annotation API to support segmentation and classification projects. You can use this to build integrations with other labeling tools or to create a Label Assist feature using predictions from a custom model or another tool.
Day 9: Project Renaming
Sometimes it's the little things in life: it's now possible to rename your projects in Roboflow. Sometimes the scope changes (or maybe you just made a typo). This basic functionality has been requested frequently, so we decided it was high time to build it.
Day 10: Label Assist for Segmentation Projects
We now support getting Label Assist predictions from segmentation models in addition to object detection models. This feature pairs extremely well with our Smart Polygon feature which uses an advanced ML model to help refine segmentation annotations.
Using your model to pre-annotate, and use Smart Polygon to refine those predictions will help you annotate much faster.
Day 11: Image Tagging
Our dataset search feature helps you better understand your dataset and intelligently expand it with the images most likely to improve your model's performance. You now have more power at your fingertips with image tagging & metadata search.
Day 12: Upload Model Weights to Roboflow (YOLOv8 Object Detection)
And the #Shipmas grand finale: support for the newly released YOLOv8 in Roboflow. We now support training YOLOv8 on your custom datasets, deploying YOLOv8 with Roboflow Deploy, and sharing your fine-tuned YOLOv8 models on Roboflow Universe.
We're excited to be kicking off 2023 with a bang, but this is only the beginning. Roboflow has a ton of new features in the hopper that will be released in the coming weeks and months.