ROBIN Competition

Datasets, Ground Truths and Metrics
For the evaluation of
object recognition and image categorisation

BERTIN Tech. and ECA (dataset #1)

Multi-class object detection with view point changes produced by Bertin Technologies and Cyberntix (ECA).

This dataset is made of colour and infrared images of both vehicles and pedestrians.The sensors used are :

  • a color Sunkwang Electronics SK-2172X camera with a cony CCD (24 bits RGB images).
  • a FLIR A20M thermal imager delivering pictures with a resolution of 320X240 pixels coded over 16 bits.

Classification Tasks

IN : list of classes to research, test images with lists of bounding boxes.
OUT : detected class of object for each bounding box.

Recognition : Discrimination of human and vehicles : Person, Motorbike, Tourism Car, Utility Car.


CYBECA_VIS_Reco1.png CYBECA_VIS_Reco2.png


CYBECA_IR_Reco1.png CYBECA_IR_Reco2.png

Posture : Discrimination of human postures : Hunkering Person, Laying Person, Standing Person.


CYBECA_VIS_Posture1.png CYBECA_VIS_Posture2.png


CYBECA_IR_Posture1.png CYBECA_IR_Posture2.png

Detection Task

IN : list of classes to research, test images.
OUT : list of detections (bounding boxes or point) for each image.

Detect : Detection of human and vehicles


CYBECA_VIS_Detect1.png CYBECA_VIS_Detect2.png


CYBECA_IR_Detect1.png CYBECA_IR_Detect2.png