choose an app to label ground truth data -凯发k8网页登录

  • documentation
main content

choose an app to label ground truth data

you can use computer vision toolbox™, automated driving toolbox™, lidar toolbox™, audio toolbox™, signal processing toolbox™, and medical imaging toolbox™ apps to label ground truth data. use this labeled data to validate or train algorithms such as image classifiers, object detectors, semantic segmentation networks, instance segmentation networks, and deep learning applications. the choice of labeling app depends on several factors, including the supported data sources, labels, and types of automation.

one key consideration is the type of data that you want to label.

  • if your data is an image collection, use the image labeler app. an image collection is an unordered set of images that can vary in size. for example, you can use the app to label images of books for training a classifier. the image labeler can also handle very large images (at least one dimension >8k).

  • if your data is a single video or image sequence, use the video labeler app. an image sequence is an ordered set of images that resembles a video. for example, you can use this app to label a video or image sequence of cars driving on a highway for training an object detector.

  • if your data includes multiple time-overlapped signals, such as videos, image sequences, or lidar signals, use the ground truth labeler (automated driving toolbox) app. for example, you can label data for a single scene captured by multiple sensors mounted on a vehicle.

  • if your data is only a lidar signal, use the (lidar toolbox). for example, you can use this app to label data captured from a point cloud sensor.

  • if your data consists of single-channel or multichannel one-dimensional signals, use the (signal processing toolbox). for example, you can label biomedical, speech, communications, or vibration data. you can also use signal labeler to perform audio-specific tasks, such as speech detection and speech-to-text transcription.

  • if your data is a 2-d medical image or image series, or a 3-d medical image volume, use the medical image labeler (medical imaging toolbox). for example, you can label computed tomography (ct) image volumes of the chest to train a semantic segmentation network.

this table summarizes the key features of the labeling apps.

labeling appdata sourceslabel supportautomationadditional features
image labeler

  • image collections

  • very large images (at least one dimension >8k)

  • rectangle regions of interest (rois)

  • projected cuboid (rois)

  • line rois

  • pixel rois

  • polygon rois

  • point rois

  • sublabels

  • attributes

  • scenes

  • built-in automation algorithms

  • custom automation algorithms

  • blocked image automation algorithms

  • view visual summary of labeled data

video labeler

  • videos

  • image sequences

  • custom image data sources

  • rectangle rois

  • projected cuboid (rois)

  • line rois

  • pixel rois

  • polygon rois

  • point rois

  • sublabels

  • attributes

  • scenes

  • built-in automation algorithms

  • custom automation algorithms

  • temporal automation algorithms

  • view visual summary of labeled data

ground truth labeler (automated driving toolbox)

  • videos

  • image sequences

  • custom image data sources

  • point cloud sequences (pcd or ply files)

  • velodyne® lidar files

  • rosbags (requires ros toolbox)

  • rectangle rois

  • projected cuboid (rois)

  • cuboid rois

  • line rois

  • pixel rois

  • polygon rois

  • point rois

  • sublabels

  • attributes

  • scenes

  • built-in automation algorithms, including vehicle and lane detection algorithms and a point cloud temporal interpolation algorithm

  • custom automation algorithms

  • temporal automation algorithms

  • multisignal automation

  • view visual summary of labeled data

  • connect external tool to app for displaying time-synchronized signals, such as lidar or can bus data

  • customize loading interface to support additional data sources

(lidar toolbox)

  • point cloud sequences (pcd or ply files)

  • velodyne lidar files

  • las/laz file sequences

  • rosbags (requires ros toolbox)

  • cuboid rois

  • attributes

  • scenes

  • built-in automation algorithms, including a lidar object tracker and point cloud temporal interpolator

  • custom automation algorithms

  • temporal automation algorithms

  • view the cuboid labels in top, side, and front views

  • save and reuse custom camera views

  • connect to external tool to display time-synchronized signals for ease of labeling, such as videos, to use as a reference while labeling

(signal processing toolbox)

  • numeric arrays, matlab® timetables, and labeledsignalset objects in the matlab workspace

  • mat-files and csv files

  • audio files (wave, ogg, flac, au, aiff, aifc, mp3, mpeg-4 aac)

  • time-based rois

  • time-based roi features

  • time-based points

  • attributes

  • attribute features

  • file-level labels

  • sublabels

  • built-in peak labeling

  • built-in feature extraction

  • custom automation algorithms

  • speech detection

  • speech-to-text transcription (requires audio toolbox extended functionality for (audio toolbox))

  • expand, collapse, and browse details of labeled data

  • view signal spectra and spectrograms

  • label rois and points using the spectrogram

  • label signals in bulk

  • use label viewer to view and compare labels

  • audio playback

  • inspect audio file information

  • export extracted features to classification learner (statistics and machine learning toolbox)

medical image labeler (medical imaging toolbox)

  • 2-d medical images and image series (dicom or nifti files)

  • 3-d medical image volume (dicom, nifti, or nrrd files)

  • pixel rois

  • built-in automation algorithms

  • custom automation algorithms

  • view 3-d medical images in the coronal, sagittal, and transverse anatomical planes

  • view 3-d medical images using customizable volume rendering

  • label multiple related images or image volumes in one app session

related topics

网站地图