DEVAOptions

Options & Configuration · AiDotNet.ComputerVision.Segmentation.Video

Configuration options for DEVA decoupled video segmentation.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → SegmentationModelOptions → DEVAOptions

For Beginners

These options configure the DEVA model. Default values follow the original paper settings.

Constructors

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

Properties

Property Summary
DEVAModelSize ModelSize { get; set; } Gets or sets which published size variant of the model to build.
bool UseGroupNormalization { get; set; } Gets or sets whether the native encoder uses group normalization instead of the research-default batch normalization.
double AssociationIoUThreshold { get; set; } Gets or sets the IoU a propagated segment and a new segment must exceed to be associated.
int MaxMemoryFrames { get; set; } Gets or sets the most frames the working memory holds; the first frame is always kept.
int MaxMissedMerges { get; set; } Gets or sets after how many consecutive unmatched integrations a propagated segment is deleted.
int MemoryTopK { get; set; } Gets or sets how many memory pixels each query pixel reads from (XMem’s top-k filtering).
int MergeEvery { get; set; } Gets or sets how often, in frames, the image-level segmentation is integrated with the propagated segments and the frame is added to memory.
int? DecoderDimension { get; set; } Gets or sets the decoder feature dimension.
int[] ChannelDimensions { get; set; } Gets or sets the four encoder-stage channel dimensions.
int[] StageDepths { get; set; } Gets or sets the number of blocks in each of the four encoder stages.

Methods

Method Summary
void Validate() Throws when a value on this instance cannot produce a working model.