1. Segment Anything#
It seems that a proxy is needed to access their official website, and they also provide a Github repository for local deployment.
2. How to Use#
Hover & Click#
First, randomly select a simple composition to test the effect.
When the mouse moves, it will automatically select the whole at the mouse position.
Clicking the left mouse button will select it. If it is not completely selected, you can select the unselected part with the left mouse button.
If you select an unnecessary part, select Remove Area on the left panel and click on the unnecessary part to remove it.
After the selection is complete, click on Cut out object on the left panel to output.
This is the output effect. It can be seen that the recognition rate is acceptable, but there is no feathering, so the edges are not very smooth.
Box#
In addition to the first click selection method similar to PS templates
It also provides a Box method, which automatically recognizes objects when the mouse is long-pressed and dragged to select.
This time we use a slightly more complex image to see the effect.
Only units that are fully selected by the selection box will be recognized, and those partially selected will not be recognized.
The overall recognition rate is okay, but if the image is too large and there are too many distractions, it will select some strange things.
Everything#
Segment Anything also provides a fully automatic way to cut out images.
This method will automatically select all objects in the image and output them as separate files.
Let's use an image with a lot of objects, and the scanning time is about 2-3 seconds.
Indeed, it cannot properly cut out images with a bunch of beans.
It can recognize white radishes properly, but the edge processing is not very good.
3. Summary#
Overall, the effect is relatively average, although the processing speed and recognition rate are acceptable.
However, as an auxiliary tool, it is already qualified, although the effect is not as good as removebg.
But it is open source and free, and I hope that future updates can improve the edge processing.