poly2d

The VisionAI format provides the polygons (2D) shape “poly2d“ to describe individual objects. They can be expressed as lists of x,y-coordinates using MODE_POLY2D_ABSOLUTE.

Example

"poly2d": [{
        "name": "poly2d_shape",
        "val": [400, 200, 450, 250, 500, 300],
        "mode": "MODE_POLY2D_ABSOLUTE",
        "closed": true,
        "stream": "camera1",
        "attributes" : {
            "boolean" : [{
                    "name" : "visible",
                    "val" : false
                }, {
                    "name" : "occluded",
                    "val" : false
                }
            ],
            "text" : [{
                    "name" : "brand",
                    "val" : "toyota"
                }, {
                    "name" : "color",
                    "val" : "red"
                }
            ]
        }
    ]
}

Schema

name	description	type	unit	required
${OBJECT_TYPE}	The name of this type. This case is “poly_2d”.	object	-	true
name	The name of this poly2d. Usually to be "poly2d_shape".	string	-	true
val	Meanings of each element in order. X-coordinate is at the element with odd order. Y-coordinate is at the even one.	even number elements array of int	px	true
closed	Indicated whether this shape is closed. If this value is true, this shape is a polygon. Otherwise, it is a polyline.	boolean		true
mode	Define the mode of the polyline list of values. Currently, we only allow MODE_POLY2D_ABSOLUTE in the field. MODE_POLY2D_ABSOLUTE means every point of this polygon is represented by its absolute coordinate.	string		true
stream	Represents which stream this shape is on.	string	-	true
confidence_score	The confidence score of model prediction of this object. Ground truth does not have this attribute.	number	-	false
attributes	attributes this bounding box has	object	-	false

<準備一張圖>

hus, a user application can directly read specific objects, without the need to load the PNG image and find the object of interest. The counterpart is an increased JSON size. Polygons (2D) can be expressed directly as lists of x,y-coordinates, using MODE_POLY2D_ABSOLUTE. However, this may create very large and redundant information. Lossless compression mechanisms can be applied to convert the, possibly long, list of x,y-coordinates into smaller strings:

---

Use Case

polygon

To describe a polygon dataset with one camera sensor:

• sensor: camera (#camera1)

• ontology:

  • people

    • ischild - boolean (static info)

    • direction - front, left, right, back (dynamic info)

    • age - number (static info)

  • car

    • color - white, silver, blue, red, black (static info)

  • truck

  • bus

Example Code

pagepolygon

polyline

To describe a polyline dataset with one camera sensor:

• sensor: camera (#camera1)

• ontology:

  • landmarkings

    • color (vec) - white, yellow, red (static)

    • type (vec) - solid, dash (static)

    • lane - number (dynamic)

  • curb

Example Code

pagepolyline