[Computer Vision] Image, Video 분야 subtask 및 데이터 종류 정리

Task 종류와 관련 dataset (핫한 순위별로)

• Video Object Tracking
  • Dataset
    • OTB(Object Tracking Benchmark)
• Video Action Classification / Action Recogntion 
  • Dataset
    • UCF101: 13,320 video clip and its categories(101 categories, divided into 5 types: 1)Human-Object Interaction 2) Body-Motion Only 3) Human-Human Interaction 4) Playing Musical Instruments 5) Sports)
    • Kinetics: 500,000 video clips(around 10 second, high-quality) and 600 human action classes
    • HMDB51: 6,849 video clips and 51 action categories
• Video Object Segmentation / Semantic Segmentation / Panoptic Segmentation
  • Dataset
    • DAVIS(Densely ANnotated Video Segmentation): 50 video sequences with densely annotated frames
    • Cityscapes: images(5000 fine annotated, 20000 coarse annotated, X video), pixel annotations for 30 classes 
    • NYUv2(NYU-Depth V2): a variety of indoor scenes as recorded by both the RGB and Depth camera
• Video Understanding
• Video Classification (not action classfication)
  • : task of producing a label that is relevant to the video given its frames
• Video Prediction
• Video Super-Resolution
• Video Compression
• Human Pose Estimation
  • Dataset
    • Human3.6M: motion capture datasets, accurate 3D joint positions(3.6M human poses) and high-resolution video
• Video Dense Estimation
  • : task of estimating depth 
  • Dataset
    • NYUv2(NYU-Depth V2)

Image dataset

Reference

• CIFAR-10
  • CIFAR-100
• ImageNet
• COCO
• MNIST
  • Fashion-MNIST
• Cityscapes
• CelebA
• CUB-200-2011
• Visual Question Answering
• SVHN