XDecoderModelSize

Enums · AiDotNet.Enums

Defines the backbone size variants for X-Decoder.

For Beginners

X-Decoder is a generalist vision decoder that handles referring segmentation, open-vocabulary segmentation, and image captioning in a single unified model. It decodes both pixel-level masks and text tokens using one shared architecture.

How It Works

Technical Details: Uses a two-path decoder: one for pixel-level predictions (masks) and one for token-level predictions (text). Both paths share the same attention mechanism. Supports various segmentation and vision-language tasks without task-specific heads.

Reference: Zou et al., “Generalized Decoding for Pixel, Image, and Language”, CVPR 2023.

Fields

Field Summary
XDecoderModelSize Base Base backbone (~86M params).
XDecoderModelSize Large Large backbone (~307M params).
XDecoderModelSize Tiny Tiny backbone (~30M params).
int value__