I’m using the SAM2 block followed by the Mask Visualization block, but SAM2 is not recognizing even basic, well-defined objects (even with several thresholds tested). As a result, no masks are overlaid on the image. Is there a link to a successful, complex workflow using SAM2 and Mask Visualization blocks that could be provided so I can fork it? Or a solution to this error? That would be greatly appreciated. I’m using a dedicated deployment with a GPU.
Hi there! When SAM2 is used without being grounded, it just returns foreground predictions, which might be none in the case of your images.
It is recommended to first run an Object Detection model (such as YOLO-World), and then use the resulting predictions in the Boxes parameter for SAM2 to turn them into polygons.
Can I ask what you are trying to detect?
I was testing SAM via their demo (https://segment-anything.com/) and it was able to identify each individual puzzle piece and apply a colored mask to each piece. How would I go about doing this for the attached image with a workflow and SAM2?
This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.