Hello, Good People.
Both are installable from Github and work out-of-the-box on the csv output from Mbaza:
1. mbaza-sequencer: Group images (sequences) together to give a combined prediction
Experiments find this improves classification accuracy by up to 5%.
In the example video attached, the delay max_delay=5, means images will be sequenced together until the image-to-image delay is > 5 seconds.
Why is it useful?
The first prediction in the csv is of a hare and you will see why when you see the first image! (shown near the end of the video).
Combining the predictions corrects this to an aardvark.
2. mbaza-mv-predicted: Move classified images to year / week / species folder
Both work on the .csv file output from Mbaza for Python 3.8+. The repositories also have installation and usage instructions, let us know if anything is unclear or missing in the documentation.
Here is a video demo on how to use the two add-ons.
A couple of things to note:
- The first prediction from the Mbaza .csv file is a Hare, but the sequencer corrects this to an Aardvark after grouping predictions (images shown at end).
- In the example used, an image-to-image delay to sequence images is used (< 5 seconds) but you can also specify a maximum number of images per sequence (see the documentation for all options)
If you have any questions, do let me know. You can reach out to me on LinkedIn or by email at marundu[at]appsilon[dot]com.
Spread the word!
Whether currently working on a project or exploring capabilities, we encourage you to explore all the good and the quality of work that can be achieved through applied data science solutions. Join the Data4Good initiative and discover how we can collaborate to actively contribute to positive change through the power of data.
Appsilon’s Data4Good offers a range of products and services aimed at driving meaningful impact, including advanced data dashboards, and computer vision and AI solutions. To find out more about Data4Good projects and how we can work together, contact the Data4Good Team.