Applied Object Detection

Real world applications tend to impose requirements in mixed aspects on object detection:

• person detection in crowded scenes
• long term person tracking from a single and multiple cameras
• densely packed product detection and one-shot learning

Dataset Curation

It is unlikely one dataset suffices for all purposes but a careful selection and training strategy work better.

General

• COCO2017
• OpenImages

Person Detection

• WiderPerson
  • Images/xxxxxx.jpg
  • Annotations/xxxxxx.jpg.txt

  #dets
  cls x1 y1 x2 y2
  

• CrowdHuman

Objects in the Wild

• SKU110K
  • images/[train | val | test]_xxxx.jpg
  • annotations/annotations_[train | vak | test].csv

  filename,x1,y1,x2,y2,label,width,height
  val_0.jpg,5,1429,219,1612,object,2336,4160    
  

• Object365: academic use only