Source: Deep Learning on Medium
We want to thanks you guys for all the feedback over the last few months. A lot of new cool features were developed because of your help, but we haven’t managed to share those updates with you. It’s time to fix it!
So here are some top features that you probably haven’t heard of yet:
1. Key-points aka Landmarks tool
A long requested feature is finally here!
Supervisely already has a variety of tools to deal with labeling: rectangles, polygons, polylines, brush and even Smart Tool. But what if you need to annotate a skeleton of a person? Or label a keypoints of a face?
Now you can define a custom shape (like skeleton, or face landmarks) in a special interface and use it as a quick template in our labeling interface with a new “Keypoints” tool.
- Attach custom label to a node or an edge
- Select custom colors or nodes or edges
- Remove nodes that are not presented in the image
- Resize the whole template to match an object or move points individualy
We hope that with new keypoints tool creating a complex labeling will become as simple as never before!
You can read more about this instrument in our blog post.
2. Python Scripts
Some time ago we released Python Notebooks: an intergrated Jupyter server containers that let you code cool stuff like visualizations or data manipulations with help of our SDK right in your browser.
Python Notebooks are good, but when it comes to a repetitive operations like data preparation for training or even downloading your project, it’s not convenient to run a whole Jupyter serverjust to execute a simple code.
That’s why we introduce Python Scripts: just like DTL, but with Python!
Select “Run Python Script” in project’s context menu, select one of predefined scripts, enter parameters and click “Run Script” — no need to write code or deal with complex JSON schemas.
New “Download as” options already use “Python Scripts” instead of DTL — but you can always switch back if needed: just look for “Run DTL → Download as” menu.
Soon we will add an option to write custom scripts, put them in context menu and run it on your own computers via agents — stay tuned!
3. Remove and restore Workspaces
Now you can put the whole Workspace in the Trash Bin and restore it if needed.
4. Erase underlying pixels via brush
When you do the labeling for an image segmentation task, it is common to have each pixel of you image be associated with exactly one class. But when you draw a new object on top other, you will end up with overlapping colors, which is not very convenient.
With new “Erase underlying pixels in other Bitmap objects” mode other bitmaps will be automatically corrected so that you will send up with no overlapping pixels.
If you found this post interesting, then let’s help others too. More people will see it if you give it some 👏.