Insect Soup Challenge!
At the ICCV 2013 International Workshop on
Computer Vision for Accelerated Bioscience
As part of the ICCV 2013 Workshop on Computer Vision for Accelerated Bioscience we are giving participants the opportunity to help tackle a significant bottleneck that exists in entomological research.
Insect traps are capable of capturing large numbers of insects at a specific location and the resulting collection is sometimes called an Insect Soup. It is relatively straight forward to photograph insect soups but, to date, the required sorting and cataloguing has often been completed by hand. The separation of insects into their ordinal groups (bee, wasp, ant) is incredibly valuable however the volume of insect soup imagery often can exceed the time an individual researcher can devote. Automation and crowd sourcing (or some mix) is potentially able to unlock a great deal of valuable scientific data from within the insect soup imagery.
Using the Insect Soup imagery supplied by the Australian Museum (below), participants are invited to answer the question “What has been captured in the trap?” Computer vision techniques are to be used to extract meaning from the imagery by categorising and reporting on specimen qualities such as:
- Colour, Size, Shape, and/or Wing Venation
- Location (via bounding box or colour mask)
- Grouping of specimens within an image into like groups approximating:
- Taxonomic Orders eg Hymenoptera, Diptera
[See Bioscience Resources section below for further information]
- Counts of:
- Overall number of specimens in an image
- Diversity of groups in an image, for example:
- Of Orders – number of different Orders in an image
- Of morphospecies – number of different morphospecies in an image
- Abundance within each group in an image, for example:
- Number of individual specimens in each order in an image
- Number of individual specimens in each morphospecies in an image
- Similarity measures within groups
Automated categorisation can be either supervised (with sensible training examples chosen from within the datasets) or unsupervised (with exemplar specimen images provided by the algorithm for later human identification).
This is by no means a complete list and participants are encouraged to report on any other interesting information they gleam from the provided data sets.
Entry is open to anyone, from researchers to students, enthusiasts, and professionals, other than the judges and workshop organisers. Submissions may come from individuals or teams. Judging will be conducted by the workshop chairs together with Paul Flemons (Manager of Collection Informatics, Australian Museum).
If your submission is accepted, you will be invited to present it at the CVAB workshop. Live demonstrations are possible, although not required. If for any reason you are unable to attend the workshop we may consider screening a 3min video of your entry during the workshop.
Each participant in the challenge will be asked to submit a report of up to 4 pages outlining their approach and summarising their results. Submissions must be made via the workshop submission system: https://cmt.research.microsoft.com/CVAB2013/
Submissions must be formatted according to one of the ICCV 2013 2-column templates available here: http://www.iccv2013.org/author_guidelines.php#formatting
Videos are also welcome. They should have a length no longer than 3 minutes, and should include the title along with the names and affiliations of the contestants. The submitted file should be a high quality compressed video with a size of no more than 30 MB. We will accept videos in MPEG (.mpg), Quicktime (.mov), AVI (.avi), MP4 (.mp4), Matroska (.mkv), or Flash Video (.flv) formats.
Accepted submissions will be made available via the CVAB Challenge website.
Submissions due 10 Oct 2013 [Extended]
Notification of acceptance 18 Oct 2013
Submission of camera ready reports due 29 Oct 2013
Workshop 2 Dec 2013
Workshop registration is via the ICCV 2013 website. For accepted submissions (only), presenters will be offered an extended 'early-bird' deadline of October 25.
Taxonomic Orders eg Hymenoptera, Diptera
- Definition - http://www.askabiologist.org.uk/from_the_lab/species-concepts
- Example application papers
A list of some software toolkits that you might find useful can be found at:
Insect Soup Data from the Australian Museum
These Images are licensed under a Creative Commons Attribution 3.0 Unported License.
Each Thumbnail is linked to the respective full 18MB TIFF.