package delf.protos

Get desktop application:
View/edit binary Protocol Buffers messages

Used in: DelfFeature

• optional DatumShape shape = 1

• oneof kind_oneof

  • FloatList float_list = 2

DatumShape is array of dimension of the tensor.

Used in: DatumProto

• repeated int64 dim = 1

• optional string model_path = 1

  Path to DELF model.

  Required.

• repeated float image_scales = 2

  Image scales to be used.

• optional DelfLocalFeatureConfig delf_local_config = 3

  Configuration used for DELF local features.

FloatList is the container of tensor values. The tensor values are saved as a list of floating point values.

Used in: DelfFeatures

• optional DatumProto descriptor = 1

• optional float x = 2

• optional float y = 3

• optional float scale = 4

• optional float orientation = 5

• optional float strength = 6

• repeated DelfFeature feature = 1

Used in: DelfConfig

• optional bool use_pca = 1

  If PCA is to be used, this must be set to true.

• optional string layer_name = 2

  Target layer name for DELF model. This is used to obtain receptive field parameters used for localizing features with respect to the input image.

• optional float iou_threshold = 3

  Intersection over union threshold for the non-max suppression (NMS) operation. If two features overlap by at most this amount, both are kept. Otherwise, the one with largest attention score is kept. This should be a number between 0.0 (no region is selected) and 1.0 (all regions are selected and NMS is not performed).

• optional int32 max_feature_num = 4

  Maximum number of features that will be selected. The features with largest scores (eg, largest attention score if score_type is "Att") are the selected ones.

• optional float score_threshold = 5

  Threshold to be used for feature selection: no feature with score lower than this number will be selected).

• optional DelfPcaParameters pca_parameters = 6

  PCA parameters for DELF local feature. This is used only if use_pca is true.

Used in: DelfLocalFeatureConfig

• optional string mean_path = 1

  Path to PCA mean file.

  Required.

• optional string projection_matrix_path = 2

  Path to PCA matrix file.

  Required.

• optional int32 pca_dim = 3

  Dimensionality of feature after PCA.

  Required.

• optional bool use_whitening = 4

  If whitening is to be used, this must be set to true.

• optional string pca_variances_path = 5

  Path to PCA variances file, used for whitening. This is used only if use_whitening is set to true.

FloatList is the container of tensor values. The tensor values are saved as a list of floating point values.

Used in: DatumProto

• repeated float value = 1