ROBIN Competition

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

CNES (dataset #2)

Generic Object classification in satellite images produced by CNES.

The ten object classes that have been selected to be research on the SPOT 5 satelite images are:

1. Roundabouts (RO : Ronds-Points)
2. Crossroads (CR : Carrefours Routiers)
3. Highways and trunk roads (RP : Routes Principales)
4. Secondary roads (RS : Routes Secondaires)
5. Minor roads (PR : Petites Routes)
6. Tracks and ways (CH : Chemins)
7. Insulated building (BT : Bâtiments isolé)
8. Suburban Area, house gathering (ZP : Zones Pavillonnaires)
9. Bridges (PT: Ponts)
10. Railways (VF : Voies Ferrées)

Classification Tasks

IN : list of regions of interest (128x128 pixels) containing one class each.
OUT : class of the object contained in each region of interest.

C1 : Discrimination of linear structures ("Tracks", "Minor Roads","Highways", "Secondary roads" or "Railways")

tracksAndWays.png minorRoads.png Highway.png

C2 : Discrimination of compact structures ("Insulated building", "Crossroads", "Bridges" or " Roundabouts" )

crossRoad.png roundAbout.png InsulatedBuilding.png

C3 : Discrimination of unknown structures

bridge.png Housegathering.png railway.png

Detection Tasks

Local detection :
IN : list of regions of interest (128x128 pixels).
OUT : boolean = true if there is an object centered in the region of interest (at 3 pixels).

D1 : Detection of linear structures ("Tracks", "Minor Roads","Highways", "Secondary roads" or "Railways")

tracksAndWays2.png none.png Highway2.png

D2 : Detection of compact structures ("Insulated building", "Crossroads", "Bridges" or " Roundabouts" )

crossRoad2.png none2.png roundAbout2.png

D3 : Detection of unknown structures

bridge2.png Housegathering2.png none3.png

For more information about the dataset and the scenarii, please refer to this presentation.