Struct Sam2DecoderConfig 

pub struct Sam2DecoderConfig {

    pub transformer_dim: usize,
    pub transformer_depth: usize,
    pub transformer_num_heads: usize,
    pub transformer_mlp_dim: usize,
    pub num_mask_tokens: usize,
    pub iou_head_depth: usize,
    pub iou_head_hidden_dim: usize,
    pub iou_prediction_use_sigmoid: bool,
    pub use_object_pointer: bool,
    pub use_mlp_for_obj_ptr_proj: bool,
    pub pred_obj_scores: bool,
    pub pred_obj_scores_mlp: bool,
    pub use_multimask_token_for_obj_ptr: bool,
    pub use_high_res_features: bool,
    pub dynamic_multimask_via_stability: bool,
    pub dynamic_multimask_stability_delta: f32,
    pub dynamic_multimask_stability_thresh: f32,
    pub layer_norm_eps: f64,
}

Mask decoder configuration. Field names + defaults mirror sam2/modeling/sam/mask_decoder.py::MaskDecoder.__init__ and the published sam2_hiera_*.yaml model.sam_mask_decoder_extra_args.

Fields

transformer_dim: usize

transformer_depth: usize

transformer_num_heads: usize

transformer_mlp_dim: usize

num_mask_tokens: usize

4 = 1 best-mask token + 3 multimask tokens (num_multimask_outputs=3 in the YAML; total tokens = num_multimask_outputs + 1).

iou_head_depth: usize

iou_head_hidden_dim: usize

iou_prediction_use_sigmoid: bool

iou_prediction_use_sigmoid flag (true in the published YAML).

use_object_pointer: bool

SAM 2 emits an additional object-pointer token. Always true for the published video configs.

use_mlp_for_obj_ptr_proj: bool

If true, obj_ptr_proj is a 3-layer MLP; else a plain Linear. True in every published config.

pred_obj_scores: bool

Predict an object-score logit (whether an object is present). True in every published config.

pred_obj_scores_mlp: bool

If true, pred_obj_score_head is a 3-layer MLP; else a Linear. True in every published config.

use_multimask_token_for_obj_ptr: bool

When multimask is selected, use the three multimask tokens (rather than the best-mask token) for the object pointer. True in every published config.

use_high_res_features: bool

Use the FpnNeck’s stride-4 + stride-8 features to refine the upscaled mask. True in every published config.

dynamic_multimask_via_stability: bool

Fall back to the best multimask output when the single-mask token’s stability score is below threshold. True in every published video config (dynamic_multimask_via_stability=True).

dynamic_multimask_stability_delta: f32

dynamic_multimask_stability_thresh: f32

layer_norm_eps: f64

Trait Implementations

impl Clone for Sam2DecoderConfig

fn clone(&self) -> Sam2DecoderConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Sam2DecoderConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Default for Sam2DecoderConfig

fn default() -> Sam2DecoderConfig

Returns the “default value” for a type.