Description of computer vision datasets

MNIST (http://yann.lecun.com/exdb/mnist/)

• The MNIST database of handwritten digits, available from this page, has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image.

  • Handwriting digits from 0 to 9
  • 60,000 training examples features labels
  • 10,000 test examples features labels
  • Image classification

Imagenet (http://image-net.org)

• ImageNet is an image dataset organized according to the WordNet hierarchy. Each meaningful concept in WordNet, possibly described by multiple words or word phrases, is called a "synonym set" or "synset". There are more than 100,000 synsets in WordNet, majority of them are nouns (80,000+).

• Summary and Statistics (updated on April 30, 2010)

  • Total number of non-empty synsets: 21841
  • Total number of images: 14,197,122
  • Number of images with bounding box annotations: 1,034,908
  • Number of synsets with SIFT features: 1000
  • Number of images with SIFT features: 1.2 million

• Challenages

  • ILSVRC2017 results
    • Squeeze and Excitation Network (webpage, arXiv)
    • Dual Path Networks (github)
  • ILSVRC2016 results
  • ILSVRC2015 results
  • ILSVRC2014 results
  • ILSVRC2013 results
  • ILSVRC2012 results
  • ILSVRC2011 results
  • ILSVRC2010 results

COCO (http://cocodataset.org)

• COCO is a large-scale object detection, segmentation, and captioning dataset.

  • Object segmentation
  • Recognition in context
  • Superpixel stuff segmentation
  • 330K images (>200K labeled)
  • 1.5 million object instances
  • 80 object categories
  • 91 stuff categories
  • 5 captions per image
  • 250,000 people with keypoints

• Challenages

  • COCO2018 Detection task with object segmentation Ongoing

    • Use the 2017 datasets for detection and keypoint

      • Download with gsutil -m rsync gs://images.cocodataset.org/val2017 val2017, and replace val2017 with train2017, val2017, test2017, unlabeled2017
      • Train Images Train/Val Annotations 118K/18GB
      • Validate images Images Staff Train/Val Annotations 5K/1GB
      • Test Images Info 41K/6GB
      • Unlabeled Images Info 123K/19GB

    • Cocoapi for Lua, MATLAB, Python

WebVision (https://www.vision.ee.ethz.ch/webvision/index.html)

• Designed to facilitate the research on learning visual representation from noisy web data.

• WebVision Dataset 2.0 (Over 5,000 synsets and 16 million images images crawled from the Flickr website and Google Images search)

• Challenages

  • WebVision Image Classification Task 2018 Ongoing