Object Detection

These are images which are possibly of use for testing object detection strategies as discussed in the CCV book.

Set 1. Pedestrians with body direction: Bounding boxes (about 11,580 in total) located in folders named 1, 2, 3, …, or 8 contain pedestrians appearing with a body direction manually classified as ground truth into North, NorthEast, East, SouthEast, South, SouthWest, West, or NorthWest, respectively. The zip-file contains a ReadMe for further details. Junli Tao, September 2014.

Pedestrians with Body Directions (ZIP, 60 MB)

Set 2. Detection of human heads: I took those 200 photos in August 2014 in Florence, Italy. They show statutes, mannequins, and further objects in Florence which represent, or do not represent human heads. This is certainly a challenging exercise for human head detection. There are easy images (where faces have been properly detected while shooting with a simple camera), and there are very very difficult images, not expected to be solvable by automated human head detection for a long time. Labelling of true-positves has been done manually. As a copyright owner of those images I am releasing them into the public domain. Reinhard Klette, September 2014

Florence Head Detection Images (ZIP, 613 MB) Labelled Florence Head Detection Images (ZIP, 695 MB)

Set 3. Detection of human heads: I took those 281 photos in November 2014 in Singapore. They show statutes, mannequins, and further objects in Singapore which represent, or do not represent human heads. This is another challenging exercise for human head detection. Labelling of true-positves has been done manually. As a copyright owner of those images I am releasing them into the public domain. Reinhard Klette, November 2014

Singapore Head Detection Images (ZIP, 1.73 GB) Labelled Singapore Head Detection Images (ZIP, 1.61 GB)
Skip to toolbar