PerceptionEncoderOptions

Options & Configuration · AiDotNet.VisionLanguage.Encoders

Configuration options for Meta’s Perception Encoder vision foundation model.

Inheritance: ModelOptions → VisionEncoderOptions → PerceptionEncoderOptions

How It Works

Perception Encoder (Meta, 2025) is a vision encoder designed for multimodal alignment tasks. It combines contrastive learning with dense prediction objectives to produce features suitable for both global understanding and local spatial reasoning.

Constructors

Constructor Summary
PerceptionEncoderOptions() Initializes a new instance with default values.
PerceptionEncoderOptions(PerceptionEncoderOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
PoolingStrategy GlobalPooling { get; set; } Gets or sets the global feature pooling strategy.
bool UseDenseFeatures { get; set; } Gets or sets whether to use dense prediction features alongside global features.
int AlignmentProjectionDim { get; set; } Gets or sets the alignment projection dimension.
int DenseFeatureStride { get; set; } Gets or sets the dense feature output stride.