• Regarding species richness, isn't it covered by the activity tab where we can see the entire list of detected species? What else do you think would be helpful? I can imagine a better UI focusing on species with more information for each species and easier search but the raw info would be roughly the same.

I think the app sort of covers this, using the pie chart layover on leaflet in the activity tab. However, it would be nice to have a more direct way of visualizing the species richness (i.e. scale the radii of the circle markers with the number of species detected). In addition to this you may want to think about visualizing simple diversity indices (there's alpha diversity which captures the species richness at each trap, gamma diversity to summarize the total richness in the area and beta diversity to assess how different species compositions are between traps). Note: I do not use diversity indices often enough to provide more specific guidance. @ollie_wearn is this what you were referring to?

• Regarding occupancy, that makes sense to me. Only challenge is I'm not using python or R at the moment because and we use javascript for easier user interfaces with web technologies. It's still possible to add python support but I'd delay as much as possible to keep things simple. Anyway, it's a technical challenge that I'm pretty sure I can solve.

I can help you with setting up the models in R, Python or Stan/ C++. But I have no idea on how to overcome the technical challenge that you are referring to. I agree with @ollie_wearn that allowing variable selection and model building would take this to far. One thing I would suggest is to allow users to fit a spatially-explicit version of the basic occupancy model (mind that these can be slow). This type of model leverages the correlations in species detections between trap locations to estimate differences in occupancy across the study area rather than just estimating a single occupancy for the entire area. 

• What about relative abundance? My readings always mention occupancy, species richness, abundance and density. I've read and ask enough about density to know that it's hard for unmarked animals and might not be necessary. If I understood correctly, relative abundance can give similar insights and as a conservationist, you probably want to know the trend of relative abundance over time.

Yes, I would leave users with the following options: a UI-switch that let's them pick either: 

1. the number of observations
2. the number of individuals detected
3. the relative abundance index or RAI (based on 1.)
4. the RAI (based on 2.)

to visualize on the leaflet map - and barchart on the side in the activity tab.

Regarding density: you could add a tab that calculates density using the Random Encounter Model (REM), which is often used when estimating density of unmarked animals without info on recaptures.

Regarding activity patterns: I would also add a tab were users can visualize either diel or annual activity cycles (often called activity patterns) computed through the activity R-package (or integrate this in an existing tab). And maybe even allow computing overlap in daily activity cycles among selected species.

If you manage to include all of these, then I think your app covers the 90% of use cases.

Some other features worth discussing:

1. merging/ comparing multiple projects
2. in line with calculating overlap between activity cycles, allow computing a spatial overlap between two species or even a spatio-temporal overlap 

@Jeremy_ For the Python implementation of basic occupancy models (as suggested by @ollie_wearn ), please refer to these two projects:

• https://github.com/timmh/biolith
• https://github.com/kenkellner/pyunmarked

I second @martijnB suggestion to use spatially explicit occupancy models (as implemented in R, e.g., https://doserlab.com/files/spoccupancy-web/). However, this would need to be added to both of the aforementioned Python projects.

Ah I see, thanks for clarifying.

BTW yolov7 actually came out earlier than yolov6. yolov6 has higher precision and recall figures. And I noticed that in practise it was slightly better.

My suspicion is that it's not trival to translate the layer functions from yolov6 or yolov9 to hailo specific ones without affecting quality in unknown ways. If you manage to do it, do tell :)

The acceptability of a drop of performance depends heavily on the use case. In security if I get woken up 2x a night versus once in 6 months I don't care how fast it is, it's not acceptable for that use case for me.

I would imagine that for many wild traps as well a false positive would mean having to travel out and reset the trap.

But as I haven't personally dropped quantization to 8-bits I appreciate other peoples insights on the subject. Thanks for your insights.

@LukeD, I am looping in @Kamalama997 from the TRAPPER team who is working on porting MegaDetector and other models to RPi with the AI HAT+. Kamil will have more specific questions.

This is great news!



I am using rather high resolution images and have just ordered some 4K (8MP) camera traps.

The standard megadetector run via Addax AI is struggling a bit with detecting relatively small animals (frame wise) although they have quite a number of pixels. This naturally follows from the resizing in megadetector.

I have noticed :

Read "MegaDetector/megadetector/detection/run_tiled_inference.py at 472460d7da7de84027282841b5b775664a4305ed · agentmorris/MegaDetector · GitHub" on GitHub

MegaDetector/megadetector/detection/run_tiled_inference.py at 472460d7da7de84027282841b5b775664a4305ed · agentmorris/MegaDetector · GitHub

MegaDetector is an AI model that helps conservation folks spend less time doing boring things with camera trap images. - MegaDetector/megadetector/detection/run_tiled_inference.py at 472460d7da7de84027282841b5b775664a4305ed · agentmorris/MegaDetector

GitHub

but this seem not readilly available in Addax AI. Is it somehow supported in SpeciesNet?

Cheers,

Lars

Hi Ștefan! 

In my current case, I am trying to detect and count Arctic fox pups. Unfortunately, Arctic fox does not seem to be included in the training data of SpeciesNet but even if it was, pups look quite different from adults. 

After a quick correspondance with Dan Morris and Peter van Lunteren on the Addax AI gitHub I was made aware of the image size option of MegeDetector. It seem to help somewhat to run the detection at full resolution (in my case up to 1920*1080). I have the impression that I get more good detections, but also less false detections (even without repeat_detection_elimination) by using higher resolution.

Dan offered to have a look at my specific challenge so I sent him 10K+ images with fox pups.

Thanks for the insight Kim! It's awesome what you are doing! I am excited for any updates. 

My collegues who are looking into getting a customised 2040 DOC AI Thermal camera need something that has the battery life to be left in the field for weeks due to the remoteness of the survey sites.

Actually, detection of ~2000 "global" species is now supported in AddaxAI via supoort of SpeciesNet.

That's great!

This is more than the supply chain though. The point was the company itself cannot use any tech for anything from the 5x companies. So in my case my ISP is incompatible. Essentially I see the only companies making that kind of sacrifice are ones that want to devote themselves to defence only.

Of course. That’s US defense as a customer. European defence is fully on the table.

It’s just sad that it’s not restricted to defence. US government wildlife organisations cannot buy European tech unless that European company was pure in their eyes.

Really nice to see you put so much effort in for the birds Graeme. Top!

Yes!  

I actually have a prototype up.  Short summary is that it does most of the things I wanted it to do, but consumes too much power in the process.  

Longer story: The prototype demonstrates the following tech:

• Looping video which allows “negative trigger delay” 
• Daytime trigger via camera sensor
• Nighttime trigger via built-in LIDAR sensor
• Trigger based on MegaDetector 6c
• Saves video into iPhone video library

Unfortunately, as it stands (and as was noted earlier in this thread), it consumes too much power while doing all this.  The problem, BTW, is not the AI stuff, or even the active LIDAR sensor – it’s keeping the basic video processing pipeline -- which is necessary for the "negative trigger delay", and the AI-based trigger mechanism.    

This is definitely in the “proof of concept” and debug phase, and lacks such “niceties” as robustness, UI design, etc.  It’s very far away from the Apple store .  

I’ve taken a break from the project to contemplate options for power reduction.  I’ve been meaning to cleanup and make public the GitHub site with the prototype code, as well as some documentation.  If there’s interest in this group, especially by anyone with time/interest to look more closely at the power problem, let me know, and I’ll move it up my priority list 

Thanks Chris,

Probably quite similar in size to your existing setup above, but we'd use two Li-ion rechargable batteries most likely (could be an 18500). I'll be sure to share more information later this year.

Cheers,

Alasdair

Hi all,

I'm interested in this post for another context, tropical rainforest. I want to monitor forest cleareance in the Congo bassin rainforest through time-lapse images. Unfortunately, the Bushnell camera I used got stolen, and so I'm looking for a tiny time-lapse camera that will be more difficult to detect.

@chris_latty The picture you shared in comment looks promising. How did you do it? from a Brinno camera? Thanks.

The magic of the WILDLABS community! @Chelsea_Smith @dmorris @pvlun  thanks for showing everyone how it's done :) Chelsea, would be great to hear how this works out for you whenever you can share an update!