Data Exploration Tools provide deep understanding of your data and are crucial for building high quality models (better you understand data, more accurate models are).
This app generates report with detailed statistics for objects (Bitmap
/ Rectangle
/ Polygon
, objects of other shapes are ignored) in images project. It allows to see big picture as well as shed light on hidden patterns and edge cases (see How to use section).
Go to "Context Menu" (images project or dataset) -> "Report" -> "Object Size Stats"
If number of objects in project is huge, define random sample (%). Subset stats is meant to be an unbiased representation of entire data and will be calculated much faster. Choose the percentage of images that should be randomly sampled. By default all images will be used. And then press "Run" button. In advanced settings you can change agent that will host the app and change version (latest available version is used by default).
Once app is started, new task appear in workspace tasks. Monitor progress from both "Tasks" list and from application page. To open report in a new tab click "Open" button.
Even if app is finished, you can always use it as a history: open it from tasks list in Read Only
mode to check Input project, statistics and Output report path.
Shows input project (clickable), sample percent that user defined at start and the number of images that used in report calculation.
Shows progress bar and then the path to the saved report in Files
. So you can open it later. App saves resulting report to "Files": /reports/objects_stats/{USER_LOGIN}/{WORKSPACE_NAME}/{PROJECT_NAME}.lnk
.
To open report file from Files
use "Right mouse click" -> "Open".
Size properties for every object
Columns:
OBJECT_ID
- object idCLASS
- object class nameIMAGE
- name of the image (clickable URL) on which this object isDATASET
- dataset nameIMAGE SIZE (HW)
- image resolution in pixels (height * width)HEIGHT (PX)
- object height in pixelsHEIGHT (%)
- object height (percentage of image height)WIDTH (PX)
- object width in pixelsWIDTH (%)
- object width (percentage of image width)AREA (PX)
- object area in pixelsAREA (%)
- object area (percentage of image area)
Properties of object (in data sample) for every class. If sample == 100% then all objects are processed. Use horizontal scroll to see all columns.
Columns:
CLASS NAME
- class nameOBJECTS COUNT
- number of objects of the classMIN H (PX)
- minimum object height (in pixels)MIN H (%)
- minimum object height (in percent of image height)MAX H (PX)
- maximum object height (in pixels)MAX H (%)
- maximum object height (in percent of image height)AVG H (PX)
- average object height (in pixels)AVG H (%)
- average object height (in percent of image height)MIN W (PX)
- minimum object width (in pixels)MIN W (%)
- minimum object width (in percent of image width)MAX W (PX)
- maximum object width (in pixels)MAX W (%)
- maximum object width (in percent of image width)AVG W (PX)
- average object width (in pixels)AVG W (%)
- average object width (in percent of image width)MIN AREA (%)
- minimum object area (in percent of image area)MAX AREA (%)
- maximum object area (in percent of image area)AVG AREA (%)
- average object area (in percent of image area)
Histograms shows distributions for all classes at once. Also you can double-click to class to hide other classes except this one. Or you can select several classes to compare their distributions.
Height (in px) distribution for all classes:
Height (in px) distribution for single class:
Width (in px) distribution for all classes:
Width (in px) distribution for single class: