Feature #2491

Type for Multipie facial landmarks

Added by V. Richter about 8 years ago. Updated about 8 years ago.

Status:ResolvedStart date:02/16/2016
Priority:NormalDue date:
Assignee:-% Done:

100%

Category:-
Target version:-

Description

I would like to suggest a new ptoto-type for Multipie-Facial-Landmarks

My first idea would be to just send all points or publish the different face parts as individual repeated fields:

package rst.vision;

import "rst/math/Vec2DInt.proto";

option java_outer_classname = "FaceLandmarksType";

message FaceLandmarks {

    // @constraint(len(value) == 68)
    repeated rst.math.Vec2DInt points  = 1;
}
package rst.vision;

import "rst/math/Vec2DInt.proto";

option java_outer_classname = "FaceLandmarksType";

message FaceLandmarks {

    // @constraint(len(value) == 17)
    repeated rst.math.Vec2DInt jaw  = 1;

    // @constraint(len(value) == 4)
    repeated rst.math.Vec2DInt nose = 2;

    // @constraint(len(value) == 5)
    repeated rst.math.Vec2DInt nose_wings = 3;

    // @constraint(len(value) == 5)
    repeated rst.math.Vec2DInt right_brow = 4;

    // @constraint(len(value) == 5)
    repeated rst.math.Vec2DInt left_brow = 5;

    // @constraint(len(value) == 6)
    repeated rst.math.Vec2DInt right_eye = 6;

    // @constraint(len(value) == 6)
    repeated rst.math.Vec2DInt left_eye = 7;

    // @constraint(len(value) == 12)
    repeated rst.math.Vec2DInt outer_lips = 8;

    // @constraint(len(value) == 8)
    repeated rst.math.Vec2DInt inner_lips = 9;
}

Would you recommend the more readable version or the one that conforms with the original description?

FaceLandmarks.proto (3.29 KB) V. Richter, 02/16/2016 05:04 PM

Associated revisions

Revision c3f17498
Added by V. Richter about 8 years ago

Added face landmarks type in rst.vision

fixes #2491

  • proto/sandbox/rst/vision/FaceLandmarks.proto: new file

Signed-off-by: Jan Moringen <>

History

#1 Updated by J. Wienke about 8 years ago

Can it happen that certain points are not detected in some images?

#2 Updated by V. Richter about 8 years ago

As I understand the documentation either all points are fit more or less correct or the whole detection fails.

I attach an updated proto with documentation.

#3 Updated by V. Richter about 8 years ago

  • Status changed from New to Resolved
  • % Done changed from 0 to 100

Also available in: Atom PDF