Скачать или смотреть (Code Below) 𝐒𝐮𝐩𝐞𝐫𝐟𝐚𝐬𝐭 𝐒𝐞𝐠𝐦𝐞𝐧𝐭 𝐀𝐧𝐲𝐭𝐡𝐢𝐧𝐠 𝐨𝐧 𝐭𝐡𝐢𝐬 𝐅𝐑𝐄𝐄 𝐋𝐈𝐕𝐄 𝐀𝐩𝐩 𝐢𝐧 𝐲𝐨𝐮𝐫 𝐨𝐰𝐧 𝐬𝐲𝐬𝐭𝐞𝐦 𝐥𝐢𝐤𝐞 𝐌𝐞𝐭𝐚 𝐝𝐨𝐞𝐬

(Code Below) I have built this App to help my Medical Professionals collaborators edit and annotate masks superfast. Here is a simple, clever engineering that improves the live results dramatically.

Meta employs some clever engineering in its online version, which is only briefly mentioned in the image predictor IPython notebook file. Let me tell you the secret. "Predict with SAM2ImagePredictor.predict. The model returns masks, quality predictions for those masks, and low-resolution mask logits that can be passed to the next iteration of prediction." In this work, I use the same method with the prompts like Positive, Negative Points, along with the Bounding Box. This has been my dream project from the time I started using SAM.


SAM stands for Segment Anything. This is a foundation model by Meta. I have been using it for my medical image segmentation, and it has been extremely powerful for generating initial masks, on which my medical professional collaborators can add/remove wrong annotations. It reduces their work by 10x, I believe. However, SAM2 is not available offline in our own system - the live app system is extremely powerful because the prompts can be adjusted based on the live outputs that it gives. Now, you can do it in your own system.

Very happy today. Thanks to Meta for this beautiful work. The code is in the comment.

Very happy to write this code: https://github.com/mukherjeesrijit/sa... - I believe it will be helpful for many. I have used this for Brain Tumour, Volumetric Segmentation, Histopathology Annotation, and more.