LABIC - Bioinformatics and Computational Intelligence Laboratory

Local Repository of Research Datasets

UTFPR-CMMD: (Brazilian) Car Make and Model Dataset

1. Motivation:

The number of surveillance cameras installed in streets and roads has grown enormously during the past years due to the need to increase traffic control and security. This fact leaded to huge amounts of raw data. The identification of vehicles in these data may be of great interest to the traffic authorities or to the police. This work aims at identifying the car make, model, approximate year of fabrication and color of Brazilian cars, vans and small trucks. 

Raw data was collected by means of web scrapping. This procedure leaded to approximatelly 100000 images of 247 models of 29 car makers. After, several procedures for filtering the images was done, as follows:

• Removing non-JPEG images
• Removing images other than cars/vans/small trucks together with framing the object
• Removing images of small size or low quality
• Removing images of vehicles not belonging to the class or images with low quality due to illumination or contrast
• Reclassify images of different generations of car models

2. Dataset Description:

Preliminary version UTFPR-CMMD-V1 has 24319 images distributes into 10 makes and 50 models, between those most-sold cars in 2018 in Brazil.

More details will be available soon.

3. Link to the dataset: 

UTFPR-CMMD-V2 (dataset available soon!)

5. Related Papers:

•  Albini, L.A., Gutoski, M., Lopes, H.S., Car Make and Model Classification with Deep Learning Methods. In: Proc. Congresso Brasileiro de Inteligência Computacional - CBIC, 2019.