ALIGNOptions
Options & Configuration · AiDotNet.VisionLanguage.Encoders
Configuration options for the ALIGN (A Large-scale ImaGe and Noisy-text embedding) model.
Inheritance: ModelOptions → ContrastiveEncoderOptions → ALIGNOptions
For Beginners
ALIGN is similar to CLIP but uses a different image processing backbone (EfficientNet instead of ViT) and was trained on a much larger but noisier dataset. The key insight is that scale can compensate for noise in training data.
How It Works
ALIGN (Jia et al., ICML 2021) demonstrates that a simple dual-encoder contrastive model can achieve strong vision-language alignment when trained on a massive noisy dataset of 1.8 billion image-alt-text pairs from the web. Unlike CLIP which uses a ViT, ALIGN uses an EfficientNet as its vision encoder.
Constructors
| Constructor | Summary |
|---|---|
ALIGNOptions() |
Initializes default ALIGN options. |
ALIGNOptions(ALIGNOptions 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 UseSqueezeExcitation { get; set; } |
Gets or sets whether to use squeeze-and-excitation attention blocks in EfficientNet. |
double EfficientNetCompoundCoefficient { get; set; } |
Gets or sets the EfficientNet compound scaling coefficient. |