BiomedCLIPOptions
Options & Configuration · AiDotNet.VisionLanguage.Encoders
Configuration options for the BiomedCLIP model.
Inheritance: ModelOptions → ContrastiveEncoderOptions → BiomedCLIPOptions
For Beginners
BiomedCLIP is a version of CLIP trained specifically on medical and biological images and their descriptions from scientific papers. This means it understands medical images (X-rays, microscopy, etc.) much better than general-purpose CLIP.
How It Works
BiomedCLIP (Zhang et al., 2023) from Microsoft is a CLIP model fine-tuned on PMC-15M, a dataset of 15 million biomedical image-text pairs from PubMed Central. It uses a ViT-B/16 vision encoder and PubMedBERT text encoder, achieving state-of-the-art zero-shot biomedical image classification.
Constructors
| Constructor | Summary |
|---|---|
BiomedCLIPOptions() |
Initializes default BiomedCLIP options. |
BiomedCLIPOptions(BiomedCLIPOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
ContrastiveLossType LossType { get; set; } |
Gets or sets the contrastive loss type. |
DomainSpecialization Domain { get; set; } |
Gets or sets the domain specialization. |
PretrainingDataset Dataset { get; set; } |
Gets or sets the pre-training dataset. |
bool UseBiomedicalAugmentations { get; set; } |
Gets or sets whether to use domain-specific image augmentations. |
string MedicalTextEncoder { get; set; } |
Gets or sets the medical text encoder variant. |