Datasets for image segmentation

last modified : 04-02-2020

Here are listed all the datasets that can be used for image segmentation.

The Berkeley BBD100K

  • Contains:

    • Vehicles: Bus, Light, Sign, Person, Bike, Truck, Motor, Car, Train, Rider
    • Weather: clear, partly cloudy, over-cast, rainy, snowy, foggy, dawn/dusk, daytime, night
    • Different level of occlusion
    • Segmentation
    • Different scenes, such as: residential, highway, city, street, ...
    • Lane marking

  • Size:

    • 100,000 HD video sequences of over 1,100-hour driving experience;
    • 2D Bounding Boxes annotated on 100,000 images;
    • Segmentation over 10,000 diverse images with pixel-level and rich instance-level annotations;
    • Multiple types of lane marking annotations on 100,000 images.

  • Other details:

    • location: Different location in the USA, New York, Berkeley, San Francisco

  • Article:

    • Title: BDD100K: A Diverse Driving Video Database withScalable Annotation Tooling
    • Authors: Fisher Yu, Wenqi Xian, Yingying Chen, Fangchen Liu, Mike Liao, Vashisht Madhavan, Trevor Darrell
    • Link: article

  • Dataset: here

Cityscapes Dataset

  • Contains:

    • flat: road, sidewalk, parking, rail track
    • human: person, rider
    • vehicle: car, truck, bus, on rails, motorcycle, bicycle, caravan, trailer
    • construction: building, wall, fence, guard rail, bridge, tunnel
    • object: pole, pole group, traffic sign, traffic light
    • nature: vegetation, terrain
    • sky: sky
    • void: ground, dynamic, static

  • Size:

    • 5 000 annotated images with fine annotations
    • 20 000 annotated images with coarse annotations

  • Other details:

    • 50 cities
    • Several months (spring, summer, fall)
    • Daytime
    • Good/medium weather conditions
    • Manually selected frames:
      • Large number of dynamic objects
      • Varying scene layout
      • Varying background

  • Article:

    • Title: The Cityscapes Dataset for Semantic Urban Scene Understanding
    • Authors: Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, Bernt Schiele
    • Link: article

  • Dataset: here

The NYU Dataset

  • Contains:

    • different rooms such as: Basements, Bedrooms, Home Offices, Bathrooms, ...

  • Size:

    • 1449 densely labeled pairs of aligned RGB and depth images
    • 464 new scenes taken from 3 cities
    • 26 scene types
    • 407,024 new unlabeled frames
    • 1000+ Classes
    • Inpainted and raw depth available
    • Both object and instance labels

  • Article:

    • Title: Indoor Segmentation and Support Inference from RGBD Images
    • Authors: Nathan Silberman, Pushmeet Kohli, Derek Hoiem, Rob Fergus
    • Link: article

  • Dataset: here

Appoloscape

  • Contains:

    • Others others, rover
    • Sky: sky
    • Movable Object: car, car_groups, motorbicycle, motorbicycle_group, bicycle, bicycle_group, person, person_group, rider, rider_group, truck, truck_group, bus, bus_group, tricycle, tricycle_group
    • Flat: road, siderwalk
    • Road obstacles: traffic_cone, road_pile, fence
    • Roadside objects: traffic_light
    • Void: pole, traffic_sign, wall, dustbin, billboard
    • Building: building, bridge, tunnel, overpass
    • Natural: vegatation

  • Size:

    • It is expected that the released dataset will include 200K image frames
    • On April 03, 2018，the Scene Parsing data set cumulatively provides 146,997 frames

  • Other details:

    • Resolution: 3384 x 2710
    • Other: pixel-level annotations and pose information，depth maps

  • Article:

    • Title: The ApolloScape Open Dataset for Autonomous Driving and its Application
    • Authors: Xinyu Huang, Peng Wang, Xinjing Cheng, Dingfu Zhou, Qichuan Geng, Ruigang Yang
    • Link: article

  • Dataset: here

Playing For Data (Generated from GTA V)

  • Contains:

  • Size:

    • 24966 densely labelled frames

  • Other details:

  • Article:

    • Title: Playing for Data: Ground Truth from Computer Games
    • Authors: Stephan Richter, Vibhav Vineet, Stefan Roth, Vladlen Koltun
    • Link: article

  • Dataset: here

VKitti

  • Contains:

    • Vehicles: Cars, Bus, Van, Other
    • Weather: cloudy, sunny, rainy, night
    • Different level of occlusion
    • ...

  • Size:

    • more than 140 thousand frames
    • 8250 vehicles manually annotated
    • 1.21 million labeled bounding boxes of objects
    • ...

  • Other details:

    • location: 24 different locations at Beijing and Tianjin in China
    • 10 hours of videos captured with a Cannon EOS 550D camera
    • ...

  • Article:

  • Book:

    • Authors: ...
    • pages: ...

  • Dataset: here