ViTVariant

Enums · AiDotNet.VisionLanguage.Encoders

Specifies the Vision Transformer (ViT) variant used as the image encoder.

Fields

Field Summary
ViTVariant CoAtNet CoAtNet: Hybrid CNN-Transformer used by BASIC.
ViTVariant EfficientNetB7 EfficientNet-B7: CNN backbone used by ALIGN.
ViTVariant ViTB16 ViT-B/16: Base model with 16x16 patch size (86M params, better than B/32).
ViTVariant ViTB32 ViT-B/32: Base model with 32x32 patch size (86M params, fastest).
ViTVariant ViTBigG14 ViT-bigG/14: Biggest model with 14x14 patch size (2.54B params).
ViTVariant ViTE14 ViT-e/14: EVA-CLIP extra-large model (4.4B params).
ViTVariant ViTG14 ViT-G/14: Giant model with 14x14 patch size (1.01B params).
ViTVariant ViTH14 ViT-H/14: Huge model with 14x14 patch size (632M params).
ViTVariant ViTL14 ViT-L/14: Large model with 14x14 patch size (304M params).
ViTVariant ViTL14At336 ViT-L/14@336: Large model at 336px resolution.
ViTVariant ViTSO400M14 ViT-SO400M/14: SigLIP Shape-Optimized 400M param model.
int value__