BASICOptions
Options & Configuration · AiDotNet.VisionLanguage.Encoders
Configuration options for the BASIC (Batch-wise Alignment of Scaled Image-text Contrastive) model.
Inheritance: ModelOptions → ContrastiveEncoderOptions → BASICOptions
For Beginners
BASIC is a scaled-up version of CLIP that uses a hybrid CNN-Transformer for images (combining the strengths of both architectures) and was trained on an even larger dataset. The name “BASIC” highlights that even a basic contrastive approach works extremely well at sufficient scale.
How It Works
BASIC (Pham et al., 2022) scales up the dual-encoder contrastive learning paradigm using a CoAtNet (hybrid CNN-Transformer) as the vision encoder and a large text transformer, training on 6.6 billion image-text pairs. It achieves 85.7% zero-shot accuracy on ImageNet.
Constructors
| Constructor | Summary |
|---|---|
BASICOptions() |
Initializes default BASIC options. |
BASICOptions(BASICOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
ContrastiveLossType LossType { get; set; } |
Gets or sets the contrastive loss type. |
bool UseGradientCheckpointing { get; set; } |
Gets or sets whether to use gradient checkpointing for memory-efficient training. |
int CoAtNetVariant { get; set; } |
Gets or sets the CoAtNet variant to use as the vision encoder. |