Element

The target of the search query

The search element represents the target of the search query. This can be a class, an annotation, or a specific dataset. Search elements can be combined with other search elements by logic to refine the search query.

Can image like the query block equals to the where-clause of SQL.

Dataset

key	value	type	element logic
dataset	The dataset name.	string	= != IN ()

<Example> Find all the data in dataset 001

dataset = 'dataset 001'

# Expression 1
dataset IN ('dataset 001', 'dataset 002')

# Expression 2
dataset = 'dataset 001' OR dataset = 'dataset 002'

---

Dataslice

key	value	type	element logic
dataslice	The dataslice name.	string	= != IN ()

<Example> Find all the data in dataslice 001 or data slice 002

dataslice = 'dataslice 001'

# Expression 1
dataslice IN ('dataslice 001', 'dataslice 002')

# Expression 2
dataslice = 'dataslice 001' OR dataslice = 'dataslice 002'

---

IQA

The image quality assessment.

There are 7 types of IQA items: lightness, contrast, colorfulness, max color Kurtosis, BRISQUE, Laplacian-blur, and FFT-Blur.sql

key	value	type	element logic
iqa.lightness	The lightness of the image.	number	= > >= < <= != BETWEEN
iqa.contrast	The contrast of the image.	number	as above
iqa.colorfulness	The colorfulness of the image.	number	as above
iqa.max_color_kurtosis	The max_color_kurtosis of the image.	number	as above
iqa.brisque	The brisque of the image.	number	as above
iqa.laplacian_blur	The laplacian_blur of the image.	number	as above
iqa.fft_blur	The fft_blur of the image.	number	as above

<Example> Find all the data iqa lightness > 30

iqa.lightness > 30

---

Class

The class follows the ontology settings. In the context of your search syntax, the element “class.name” specifically refers to the classification of an object (ex.bounding box) in an image or a set of images.

Note

The class.name does not indicate whether an image or a dataset contains a certain class of objects as a whole. Instead, it is used to find specific instances of objects within the bounding boxes in the data.

key	value	type	element logic
class.name	The class name.	string	= != IN ()

<Example> Find all the data with object class car.

class.name = "car"

# Expression 1
class.name IN ('car', 'people')

# Expression 2
class.name = 'car' OR class.name = 'people'

# Expression 3
class.name = 'car' and class.name = 'people' 　# Note: There will be no result. A bounding box cannot be classified as both a car and a truck.

Hint

Use 'class IS NULL' in your search query to find items that do not have a specified object class.

---

Class with Attribute

The class attributes follow the ontology settings.

key	value	type	element logic
class.attribute.name	The name of the attribute.	string	= != IN OR AND
class.attribute.value	The value of the attribute.	Depends on the tag type - vec : string - number : number - boolean : boolean text type not included in the query.	vec, text and boolean: = != number: = > >= < <= !=

<Example> Find the person age between 28~30, year in 1995, 1996, 1997.

# search class is person who's attribute contains age that is between 28 - 30
(class.name = 'person' AND class.attribute.name='age' AND (class.attribute.value BETWEEN 28 AND 30))

# search all datarows whos attribute year is either 1995, 1996 or 1997
class.attribute.name = 'year' AND class.attribute.value IN (1995, 1996, 1997)

# search class is person or car where both attribute names contains year
(class.name = 'person' AND class.attribute.name = 'year') AND (class.name = 'car' AND class.attribute.name 'year')

# search class is person or (class is car but containes year attribute)
class.name = 'person' OR (class.name = car AND class.attribute.name = year)

---

Annotation

The target annotations result. It can be ground truth or model predictions.

key	value	type	element logic
annotation	Ground Truth or the model name.	string	= != IN ()
annotation.confidence	The confidence score of the object.	number	= > >= < <= !=

<Example> Find all the data with ground truth, model-a annotations.

annotation = 'ground_truth'
annotation = 'model-a'

# Expression 1
annotation IN ('ground_truth', 'model-a')

# Expression 2
annotation = 'ground_truth' OR dataset = 'model-a'

<Example> Find all the data with confidence score > 50% annotations.

annotation.confidence > 0.5

---

Object

The object information includes height, width, length, area, and the cuboid object distance and the points included.

key	value	type	unit	element logic
object_size.height	The object size (height)	number	px	= > >= < <= !=
object_size.width	The object size (width)	number	px	as above
object_size.length	The object size (length)	number	px	as above
object_size.area	The object area	number	px^2	as above
object_cuboid.distance	The object cuboid distance. It is only available for point cloud data cuboids.	number	m	as above
object_cuboid.point	The object cuboid included points. It is only available for point cloud data cuboids.	number	points	as above

Example:

Find all the data with object size height < 100px.

object_size.height < 100

Find all the data with object size area < 40px^2.

object_size.area < 40

Find all the data with cuboid center distance to origin > 10m

object_cuboid.distance > 10

Find all the data with cuboid containing < 5 pcd points

object_cuboid.point < 5

---

Tag

The project taggings.

key	value	type	element logic
tag.name	The name of the tag.	string	=, In !=
tag.value	The value of the tag.	Depends on the tag type - vec : string - num : number - boolean : boolean text type not included in the query.	vec, text and boolean: = != IN () number: = > >= < <= !=

<Example> Find the weather of sunny or rainy.

tag.name = 'weather'
tag.name != 'weather'
tag.name = 'weather' and tag.value IN ('sunny', 'rainy')