Repolode
Submit
lillyguisnet

lillyguisnet/sam3-segmentation

An agent skill that lets coding agents outline, mask and count objects in photos with Meta's SAM 3 on a local NVIDIA GPU.

View on GitHub
Stars
125
Forks
12
Open issues
0
Last commit
14 days ago

What is it?

sam3-segmentation is an agent skill that gives coding agents such as Claude Code, Codex and Cursor the model, commands and rules to outline, mask and count objects in photos using Meta's SAM 3. It runs on the user's own NVIDIA GPU, and the Python environments are created outside the skill folder so they can be reused across updates and installed copies.

For quick requests, the agent can find every instance of a phrase, segment an object from a click or box, or segment everything in an image. For many small objects or parts such as kernels, seeds, cells or insects, it follows a more careful workflow with overlapping close-up crops, saved raw candidates, score checks, review sheets, recorded review decisions, boundary refinement, PNG and COCO RLE exports, and a zoomable viewer.

Outputs are reported as model proposals rather than verified ground truth, with uncertain items listed. The skill does not handle video, medical images, or picture classification and editing. A demo mode runs on generated shapes without a GPU or model.

Key features

  • 01Segments every instance of a text phrase, or objects from a click, box or example box.
  • 02Runs on a local NVIDIA GPU with Python environments kept outside the skill folder.
  • 03Supports overlapping crops and saved raw candidates for many small objects.
  • 04Provides review sheets, flags and recorded review decisions.
  • 05Exports PNG and COCO RLE files and includes a zoomable viewer.
  • 06Includes a GPU-free demo that uses generated shapes.

Use cases

  • Counting and outlining kernels, seeds or cells in close-up photos.
  • Cutting out a single object such as a dog from a photo for an agent-driven workflow.
  • Producing COCO RLE masks as draft annotations to be checked by a person.
  • Trying SAM 3 segmentation on a local GPU through a coding agent.

In collections

Claude Code tools, skills & pluginsAI coding assistants & agentsAI agent skill librariesComputer vision libraries & tools

Topics

#computer-vision#image-segmentation#sam3#agent-skills#annotation#instance-segmentation#claude-code#claude-code-plugin#codex#segment-anything

FAQ

What is sam3-segmentation?
An agent skill that lets coding agents outline, mask and count objects in photos with Meta's SAM 3 on a local NVIDIA GPU. sam3-segmentation is an agent skill that gives coding agents such as Claude Code, Codex and Cursor the model, commands and rules to outline, mask and count objects in photos using Meta's SAM 3. It runs on the user's own NVIDIA GPU, and the Python environments are created outside the skill folder so they can be reused across updates and installed copies.
Is sam3-segmentation free to use? What license does it have?
sam3-segmentation is open source under the MIT license. Check the license terms before using it in commercial products.
Is sam3-segmentation still maintained?
Actively maintained. The last commit to sam3-segmentation was on Oct 24, 2026.
What can I use sam3-segmentation for?
Counting and outlining kernels, seeds or cells in close-up photos. Cutting out a single object such as a dog from a photo for an agent-driven workflow. Producing COCO RLE masks as draft annotations to be checked by a person. Trying SAM 3 segmentation on a local GPU through a coding agent.
Where can I get sam3-segmentation?
The source code and installation instructions are on GitHub: https://github.com/lillyguisnet/sam3-segmentation

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects