Data



drive_and_act

Drive&Act: Dataset for Action Recognition in Automated Vehicles

  • 12h of video data in 29 long sequences
  • Multi-modal videos: NIR, depth and color data
  • 83 manually-annotated hierarchical activity labels

          [download-page] [paper]
spase

CLEVR Scene Graphs

  • Scene graphs of the images in the CLEVR validation and test set
  • Objects were generated in TensorFlow using SSD with a ResNet152 backbone
  • These scene graphs were used for evaluating the softpaths architecture

          [val-graphs] [test-graphs] [paper]
spase

WiSe - Slide Segmentation in the Wild

  • Annotations for 1300 slides captured during lectures
  • Fine-grained pixel-wise labels
  • 14 text, 6 image-based and 4 structural classes
  • Highly overlapping segments, i.e., multiple labels per pixel

          [download-page] [paper]
spase

SPaSe - Slide Page Segmentation

  • The first benchmark dataset for slide-page segmentation
  • Annotations for 2000 complex slides
  • Fine-grained pixel-wise labels
  • 14 text, 6 image-based and 4 structural classes
  • Highly overlapping segments i.e. multiple labels per pixel

          [download-page] [paper]
zebra_crossings

Zebra Crossings and Crossing Lines

  • We labeled all images containing zebra crossings or crossing lines in the COCO dataset
  • Pixel-wise Annotations of 1000 images
  • As an annotation tool we used https://github.com/nightrome/cocostuff
  • For more information about the COCO dataset please check http://cocodataset.org
  • The annotations can be downloaded as .mat files or in a json format

          [download_zebra] [download_clines] [download_json]
driveahead

DriveAHead

  • Head Pose Dataset captured during real driving scenarios
  • Contains frame-by-frame head pose labels obtained from a motion capture system
  • Includes 1 Million depth and infra-red images

          [download-page] [paper]
Haurilet2016_WACV

Naming TV Characters by Watching and Analyzing Dialogs

  • Person number for each name mention in the subtitles
  • We include ground truth and predictions of our approach
  • Included TV-Series: Big Bang Theory, Buffy and Lost

          [download_person_nr_gt] [download_person_nr_pred] [paper]