OMGSegModelSize

Enums · AiDotNet.Enums

Defines the backbone size variants for OMG-Seg.

For Beginners

OMG-Seg (One Model that is Great for all Segmentation) handles over 10 different segmentation tasks with a single model and only 70M trainable parameters. It uses task-specific queries to switch between tasks at inference time.

How It Works

Technical Details: Uses a shared transformer backbone with task-specific query sets. Supports image segmentation (semantic, instance, panoptic), video segmentation, open-vocabulary segmentation, interactive segmentation, and more.

Reference: Li et al., “OMG-Seg: Is One Model Good Enough For All Segmentation?”, CVPR 2024.

Fields

Field Summary
OMGSegModelSize Base Base variant (~70M trainable params).
OMGSegModelSize Large Large variant (~120M trainable params).
int value__