DepthAnythingV2ModelSize

Enums · AiDotNet.Enums

Defines the encoder size variants for Depth Anything V2.

For Beginners

Depth Anything V2 estimates how far away each pixel in an image is. It comes in three sizes built on the same design: a bigger encoder gives more accurate depth but takes longer to run. Start with Base.

How It Works

Each size selects the DINOv2 encoder the paper pairs with it, which fixes both the feature width and the number of transformer blocks.

Reference: “Depth Anything V2”, NeurIPS 2024.

Fields

Field Summary
DepthAnythingV2ModelSize Base Base variant, on ViT-B: 768 features across 12 blocks.
DepthAnythingV2ModelSize Large Large variant, on ViT-L: 1024 features across 24 blocks.
DepthAnythingV2ModelSize Small Small variant, on ViT-S: 384 features across 12 blocks.
int value__