The MASTIF datasets
This page provides the traffic sign datasets collected during our research project MASTIF. The datasets may be freely used in academic research. If you find these datasets useful, please consider citing one of the following publications:- Siniša Šegvić, Karla Brkić, Zoran Kalafatić, Axel Pinz. Exploiting temporal and spatial constraints in traffic sign detection from a moving vehicle. Machine Vision and Applications. Accepted for publication.
- Siniša Šegvić, Karla Brkić, Zoran Kalafatić, Vladimir Stanisavljević, Marko Ševrović, Damir Budimir and Ivan Dadić. A computer vision assisted geoinformation inventory for traffic infrastructure. ITSC 2010, Madeira, Portugal, September 2010 (pdf).
- Igor Bonači, Ivan Kusalić, Ivan Kovaček, Zoran Kalafatić, Siniša Šegvić. Addressing false alarms and localization inaccuracy in traffic sign detection and recognition. CVWW, Mitterberg, Austria, January 2011 (pdf).
- Karla Brkić, Axel Pinz, Siniša Šegvić, Zoran Kalafatić. Histogram-Based Description of Local State-Time Appearance. SCIA, Ystad Saltsjöbad, Sweden May 2011 (pdf).
There are three datasets (TS2009, TS2010, and TS2011), which we name with respect to the year in which they were annotated. Each sign is annotated 4-5 times at different distances from the car. The developed machine learned algorithms therefore need to be carefully assessed, in order to avoid having different instances of the same physical sign in both the training and the evaluation sets.
The dataset TS2009 has been acquired with an interlaced camera, and the corresponding signs are often significantly distorted. We have therefore used TS2009 for training in most of our experiments so far. For training our detectors we have used TS2009 as is, while our classifiers have been trained on a deinterlaced variant of that dataset.
The three datasets can be downloaded from:
-
TS2009 (around 6000 signs):
- cropped sign images: tar.gz
- note that we provide only the cropped sign images for this dataset in order to save space on the server.
-
TS2010 (around 3000 signs):
- full source images + annotations: tar.gz
-
the annotation information is contained
in the text files stored in the archive
with the extension
.seq
. - the full annotated video: wmv + annotation
-
TS2011 (around 1000 signs):
- full source images + annotations: tar.gz
-
the annotation information is contained
in the text files stored in the archive
with the extension
.seq
. -
the full annotated videos:
- wmv + annotation
- wmv + annotation
- wmv + annotation
- wmv + annotation
Please note that the model images for each of the five main categories of the Croatian signs (A-E) can be downloaded from: A, B, C, D, E.
The back-end server for storing the datasets has been kindly provided by the VPS and GSS services of the University of Zagreb University computing centre. |
For all further questions and comments, feel free to send an e-mail to .