Dan's comments about the need for technologists and conservationists to manage and share (properly annotated) data struck a chord with me, it was right at the end of the presentation.

I fired off a point into the chat "could your old background data be my background data?" or something similar, and it got me thinking...

(Firstly, sorry for clouding this issue with my simultaneous "Ian Tuna" joke)

In the context of using AI as described in EdgeImpulse example, lets say - audio.

Lets say Group A are trying to detect the roar of lions, and Group B are trying to detect the grunt of wild pigs, and lets imagine that is in roughly the same area of Africa.  I mean they could both send out teams to capture the sound of their target animal, and the sound of NOT their target animal, e.g everything else.

Well if "serengetti sounds" was a known audio track, then all you need is the unique sound of the beast in question, am I right?

One barrier that was discussed in the presentation was the "risk" or uncertainty of a payoff when talking to potential funders. It seems to me that the "risk" factor of collaborations like this, for funders, would be analogous to the investment and payoff when developing a genomics program. Relative to traditional field research, it takes a much greater investment in time, with different expertise, to create a reference genome for a species of interest and then pull SNPs, etc. Nevertheless, funders seem to recognize the potential for genomic research and genomic tools. Do you think that greater openness to “risk” in conservation genomics results from closer intellectual proximity (i.e., molecular biology vs organismal biology), or to conservation genomics being a more mature field (i.e., more examples of success), or to other factors? If intellectual proximity is a key factor, then conservation tech collaborations will continue to face this barrier, but if lack of successful examples are a key factor, then perhaps this barrier will become less of an issue for conservation tech in the future. Thoughts?

Hi. Just watched the Youtube version (I'm in the unable to watch live hemisphere) and wanted to say that was a great talk. I wish it could have gone for another hour. I liked hearing about the issues and concerns from Melanie on the biology/ecology side. Also could totally relate to what Ed was talking about from the tech side about how it's difficult to have long term collaborations since there isn't really a conservation technology career path at the moment. 
I think one of the things we're trying to do with courses like Build Your Own Datalogger is to get people involved in wildlife/conservation side more comfortable with the tech. The goal is to hopefully build a shared repository of technical knowledge that people in the wildlife community can contribute to and benefit from. 

Akiba

Thanks everybody for joining today's session! I've attached the article here that describes Pattern Matching in greater detail to respond to some of the questions from today's session. 
Also, here is a link to our Privacy Policy and our Terms of Service for the individual who had questions about GDPR. Feel free to contact us at support@rfcx.org with any other questions about the platform or Pattern Matching in general! 

Zephyr and Marconi, thanks for the great talk and your hard work that made this platform available to the public!! I'm experimenting with the pattern matching function to create some training data but have issues trying to get the "Jobs'' to run. I only have 18 files for each of these tests, but only 1 processed to 11.1% while two others stay at 0%. The internet connection was briefly interrupted when I created the first job and was fine later on. I can only Hide, not delete or re-run these jobs, so I'm not sure what to do. Any suggestions would be greatly appreciated. Thanks so much!

Hi Andrew,

Yo need to train a lightweight DNN model for bird flocks which can then be deployed on Raspberry Pi. For initial starters, you can look into the below tutorial:

 https://www.pyimagesearch.com/2017/10/16/raspberry-pi-deep-learning-object-detection-with-opencv/

Best

Piyush

Hi Wildlabbers,

Just popping in to share this very cool primer for beginners to embedded machine learning from our tutor Daniel Situnayake! If you're interested in learning more about the basics of TinyML, this is a great resource.

Find the link here!

-Ellie

So interesting, thanks for sharing! It's always really helpful for the community to hear about the trial and error that goes into fieldwork. I also never realized that heat-sensing camera traps don't work on certain animals like otters, really curious to learn more - are there strategies to deal with that problem?? Maybe someone from our camera trap group could chime in on that aspect?

Hi Ellie,

thanks for your interest. So I have two possible explanations why the camera traps I used did not release, even though I found otter droppings directly in front of my camera (and later my happy face that I must have gotten them this time even though it was not the case).

I. The Cameras I used in that specific campaign where quite old and I am not too experienced in using them so maybe I did not set them up correctly. This is something I will try to overcome with newer cameras and more training in my next field campaign whenever COVID allows me to travel again...

II. Otters are very well insulated having more than 70k of hair per cm2. I found some fabulous infrared images in a book from Irene Weinberger that showed how little heat otters actually emit and that otters only differed significantly from the background radiation at there feet and nose. So another explanation of mine is that unless these bodyparts are captured by the camera's sensor the camera won´t release.

I will try new Cameratraps the next time I will have the opportunity to do fieldwork in Portugal and will let you know how it went. Until then I will try to work my way around it using footprints :-).