In this repository we are sharing the set of pictures collected by the MosquitoAlert project since 2014 for image classification tasks.

Click here to download all the images

Image Folders

Pictures are saved in folders by years. Each folder contains two files:

1. images.tar.gz.

A compressed file with the set of pictures. Each picture is a ‘.png’ file with a six digits number file name (e.g. 012345.png).

2. imgRef.txt.

A csv file with images’ info:

  • column 01. ‘imgNmb’, image number (such that the corresponding image file name is str(imgNmb).zfill(6)+’.png’);

  • column 02. ‘imgId’, MosquitoAlert internal image id (for internal purposes);

  • column 03. ‘rprtId’, MosquitoAlert internal report id (for internal purposes);

  • column 04. ‘imgDate’, image date;

  • column 05. ‘imgTime’, image time; (this is server upload time)

  • column 06. ‘imgLon’, image location longitude;

  • column 07. ‘imgLat’, image location latitude;

  • column 08. ‘imgClass’, image class as one of the following:

    • ‘mosquito species’ (e.g. ‘Ae. albopictus’, ‘Ae. aegypti’) when reporting an observation of one of our targeted species;

    • ‘otherSpecies’, when reporting an observation of any other species;

    • ‘canNotTell’, when the picture is not good enough so as to identify the species reported;

    • ‘site’, when reporting a potential mosquito breeding-site

    • ‘otherSites’, when reporting a mosquito breeding-site that is not considered as such;

  • column 09. ‘imgLabel’, image label as one of the following:

    • if the image has been validated by an expert, the label can be one of: + ‘confirmed’; + ‘probable’ (this one applies only for targeted species classes);

    • when there is not an explicit expert validation for the image, the image is labelled as: + ‘notClassified’,

  • column 10. ‘hidden’, a flag for pictures that for some reason (e.g. offensive, conflicting) are not shown in the map

Suggested classification tasks:

  1. Broad classification: classify pictures by image class

  2. Species identification: soft binary classification is suggested (e.g. ‘Ae.albopictus’ versus NO_Ae.albopictus)

  3. Site’s classification 1: drain/not drain. (Specific labels for this task will be provided in a next release.)

  4. Site’s classification 2: drains with/without water. (Specific labels for this task will be provided in a next release.)

Notes: A report may contain several pictures (same ‘rprtId’ for different ‘imgId’). Usually the pictures are similar but in some cases they are not so (e.g. a focused picture and a blurred one, a picture of a mosquito and a picture of a bite). Experts validate and label reports based on the most informative image associated to the report. For the sake of accuracy, the most informative image is the only one that is labeled and the rest remain as ‘notClassified’. For this reason, we recommend training the classifiers only with labeled images. The set of ‘notClassified’ images could be used as a test set considering the report labels as a rough indication of the image label.


In this repository we are sharing the set of pictures collected by the mosquitoAlert project since 2014 for image classification tasks. This dataset of images is distributed to the public under license Anonymous, CC by Mosquito Alert. We would like to thank the Mosquito Alert community (anonymous citizens) who have participated year by year, making all this data collection system worth it.


Please, let us know you downloaded the files, what you are planning to do and any interesting results you might get. We will be happy to hear from you.

The Mosquito Alert team,

thanks !

Click here to download the images