# \#sam

**URL:** https://community.ultralytics.com/tag/sam/6.md

[Latest](https://community.ultralytics.com/latest.md) · [Categories](https://community.ultralytics.com/categories.md) · [Tags](https://community.ultralytics.com/tags.md)

---

## [SAM/SAM 2, how to generate feature embedings and run inference based on the embedings](https://community.ultralytics.com/t/sam-sam-2-how-to-generate-feature-embedings-and-run-inference-based-on-the-embedings/1275)

<div class="topic-metadata">

**Author:** [@Kallinteris-Andreas](https://community.ultralytics.com/u/Kallinteris-Andreas)\
**Replies:** 1\
**Last updated:** [July 26, 2025, 5:30pm UTC](https://community.ultralytics.com/t/sam-sam-2-how-to-generate-feature-embedings-and-run-inference-based-on-the-embedings/1275 "2025-07-26T17:30:23Z")

</div>

the inference process of SAM has 2 main stages, (1) Generated image embedding, (2) Generated mask based on prompt. (1) takes up most of the compute time, so it is a common practice to run (1) once and (2) multiple times…

---

## [SAM ultralytics returned (estimated) IOU scores](https://community.ultralytics.com/t/sam-ultralytics-returned-estimated-iou-scores/1243)

<div class="topic-metadata">

**Author:** [@Kallinteris-Andreas](https://community.ultralytics.com/u/Kallinteris-Andreas)\
**Replies:** 1\
**Last updated:** [July 19, 2025, 11:49pm UTC](https://community.ultralytics.com/t/sam-ultralytics-returned-estimated-iou-scores/1243 "2025-07-19T23:49:40Z")

</div>

SAM2 (and SAM) besides generating masks on inference, it also returns predicted IOU scores for each returned mask, but looking at the returned value it does not contain anything indicating IOU (Pdb) p results \[ultralyti…

---

## [Ambigueous promt with SAM, return top 3 Masks](https://community.ultralytics.com/t/ambigueous-promt-with-sam-return-top-3-masks/1240)

<div class="topic-metadata">

**Author:** [@Kallinteris-Andreas](https://community.ultralytics.com/u/Kallinteris-Andreas)\
**Replies:** 7\
**Last updated:** [July 19, 2025, 1:32pm UTC](https://community.ultralytics.com/t/ambigueous-promt-with-sam-return-top-3-masks/1240 "2025-07-19T13:32:58Z")

</div>

Is there a way to get ULTRALYTICS SAM or SAM2 to provide the top 3 masks when prompted by a single point, the same way it happens on the original paper the code appears to have this capability, but I can figure out how …

---

## [SAM2 not working with ultralytics](https://community.ultralytics.com/t/sam2-not-working-with-ultralytics/379)

<div class="topic-metadata">

**Author:** [@Santhosh\_KK](https://community.ultralytics.com/u/Santhosh_KK)\
**Replies:** 6\
**Last updated:** [October 24, 2024, 1:54pm UTC](https://community.ultralytics.com/t/sam2-not-working-with-ultralytics/379 "2024-10-24T13:54:10Z")

</div>

model = SAM(“sam2\_b.pt”) video\_path = ‘test.mp4’ results = model(video\_path) → File c:\\Users\\santhosh.kumaran\\Anaconda3\\envs\\SAM2\_ENV\\lib\\site-packages\\ultralytics\\engine\\model.py:554, in Model.predict(self, source, …

---

## [SAM2 Announced!](https://community.ultralytics.com/t/sam2-announced/145)

<div class="topic-metadata">

**Author:** [@BurhanQ](https://community.ultralytics.com/u/BurhanQ)\
**Replies:** 4\
**Last updated:** [August 1, 2024, 2:08pm UTC](https://community.ultralytics.com/t/sam2-announced/145 "2024-08-01T14:08:17Z")

</div>

In case you missed it (ICYMI) Meta announced SAM2 yesterday! :rocket: SAM 2 extends its predecessor by incorporating video segmentation capabilities, using a transformer architecture with streaming memory for real-t…