discussion
Getting behavioral data out of datasets that weren't built for it
16 June 2026 3:49pm
Burning question:There's so much monitoring data already- camera trap archives, acoustic recordings, GPS tracks - but almost all of it was collected to answer presence/absence or...
13
0
10 July 2026 12:55pm
I have tens of thousands of camera trap bycatch African mammal videos that are available for analysis to anyone who can turn them into published papers, data that is actually useful for conservation, or publicity for wildlife and conservation.
They are already manually sorted into carnivores / herbivores and the carnivores are sorted and/or tagged to species. I do not have the resources to do anything further with them.
14 July 2026 1:18am
Peter, this is a generous offer - thank you. A dataset that's already sorted carnivore/herbivore with carnivores tagged to species is a real head start, and the fact that it's bycatch from another purpose is exactly the kind of "data built for one question, useful for another" material I've been thinking about.
A few questions to figure out fit: what capture mode are the videos - continuous clips, triggered bursts, fixed intervals? And roughly how long are the clips? I'm interested in whether there's enough temporal continuity to read behavior (time budgets, activity sequences), not just presence/absence. Also curious which species are best represented, and what any reuse would look like on your end in terms of credit and terms.
I'd genuinely like to explore turning some of this into something publishable and conservation-useful. Would you be open to a direct conversation off-thread?
Warmly,
Maggie
14 July 2026 1:27am
Kim, this is great - thank you for sharing it. And to answer directly: yes, footage like this is genuinely useful for behavioral work, precisely because a continuous clip establishes a whole ecological scene rather than a single detection.
What's nice here is the range of behavior visible at once. The mother is engaged in what looks like foraging, while the juveniles are showing enrichment behaviors - exploratory, playful, curious, moving freely and not staying tethered to her. Posture and pose carry a lot of the signal too: tail position, body orientation, how attention is directed. That's the kind of thing you can only read when you have temporal continuity and enough frame to see the whole animal.
One question it raises - and this is exactly the interpretive challenge I find interesting - is whether the mother is actually foraging or "reading" the landscape through scent, which looks similar on camera but means something quite different behaviorally. Disambiguating those is where the real work is.
I'd love to hear more about your setup and how much footage like this you're generating. Continuous thermal at this quality, running for months, is a valuable stream.
Maggie
TrailCam - Browser tool for preparing trail camera observations
8 July 2026 12:44am
12 July 2026 6:08pm
I’m continuing to improve this project and would really appreciate feedback from anyone who can test it. Thank you to everyone who can help the workflow with this web app becomes much faster and more efficient.
iNat TrailCam Local — V25
13 July 2026 11:13pm
This is super cool! Thanks for sharing!
13 July 2026 11:44pm
Good evening, Mr. Jorge.
The tool is currently under development; it makes submitting data to iNaturalist much faster, since you don't have to fill out all the fields.
Camera trap recommendations
2 April 2026 11:40pm
7 July 2026 1:20am
GP A60 Review is now up. See: https://winterberrywildlife.ouroneacrefarm.com/2026/05/23/gardepro-a60-trail-camera-teardown-and-review/
10 July 2026 10:22pm
It’s worth considerig thermal cameras. They will see more, particularly small mammals at night. Here’s a Comparison of thermal and trail cameras.
13 July 2026 11:18pm
I really recommend GardePro. They are not too expensive and very resilient in the field.
List of bioacoustics software
1 July 2020 3:23pm
13 June 2026 4:56pm
Hi Tessa, thanks so much for the update!!! I love that there is an interactive website now. This is such a valuable website, I'm happy to see the updates ;) - Liz
15 June 2026 9:00am
Actually, on the subject of acoustics, the Raspberry Pi based sound localization system I developed has been running continuously since 2023 writing to a 256GB SD card :-)
https://github.com/hcfman/sbts-aru
I submitted it for addition to that list a few years ago. Should be there also I guess.
I have three of these running around my house. Off power though because I can. Actually I use one of them as a time server for all my computer equipment because it maintains microsecond time accuracy continuously.
13 July 2026 11:12pm
This is a great resource!
Welcome to WILDLABS!
29 October 2015 5:16pm
10 July 2026 12:10pm
Hello everyone,
My name is Linda Mohamed, a wildlife conservationist from Tanzania with a passion for biodiversity conservation and community-based conservation. I am excited to connect with this community, learn from your experiences, and contribute to meaningful conservation efforts. I look forward to growing, collaborating, and creating a positive impact together.
12 July 2026 10:23am
Hello from Taiwan! 👋
Dear wildlife lovers,
My name is Cathy. I'm from Taiwan, and I work at TSMC as a software product manager.
Sustainability was something my schools in Taiwan started teaching us early on — the simple but powerful idea that we only have one Earth. That interest stayed with me, and I went on to study environmental economics in graduate school.
Life then took me into the semiconductor industry, and somewhere along the way that original commitment quietly moved to the back of my mind. But I recently travelled to Kenya and spent time watching the animals there, and it brought all of it rushing back — the reason I cared in the first place, and the uncomfortable realisation that I haven't yet put my skills to work for the planet in any real way.
That's why I'm here. The area I'm most drawn to is data management — how field data gets collected, structured, stored, and actually turned into something people can use and trust. In my day job I work on systems that handle large volumes of manufacturing data, and I suspect a lot of the underlying problems are the same: messy inputs, inconsistent standards, and hard-won data that never quite reaches the people who need it.
I'd love to learn what data management really looks like in conservation, and where someone with a product and systems background could be useful. If there's a project, a working group, or even just a conversation I could listen in on, I'd be very glad to hear about it.
Looking forward to learning from all of you.
Warmly,
Cathy
13 July 2026 11:10pm
Hello, I am Jorge! I have been around this community for a while, but never been here. Good to be part of the group!
Unlock New Features by Earning Your Community Involvement Badge!
14 March 2024 11:54am
5 July 2026 6:33pm
Ok
5 July 2026 6:34pm
The process begins 😂
13 July 2026 11:02pm
Awesome!
Help us protect jaguars and restore the Cerrado
13 July 2026 7:34pm
Machine Learning Engineer / Data Scientist
13 July 2026 4:24pm
Senior Coordinator, Fisheries Technology at Conservation International
13 July 2026 12:58pm
Camera Trap Suggestions for Time-lapse Seabird Monitoring?
9 July 2026 10:13pm
10 July 2026 11:35pm
Most trail cameras (rugged, small, weather resistant) have a “timelapse” mode, however, not all will do a full 24-hour timelapse of photos.
The factory firmware for the Browning trail cameras we use only allows timelapse during the day (after sunrise, before sunset).
I hacked the firmware on Browning Advantage, Edge, Elite HP4, and Elite HP5 ReconForce and SpecOps cameras so that they have an “All Day/Night” timelapse mode (including the flash at night). The PIR sensor is active even during timelapse, so the camera will also trigger on motion. If you don’t want this, you can put a piece of tape over the PIR sensor.
See: https://winterberrywildlife.ouroneacrefarm.com/2024/07/14/timelapse-feature-enhancements-for-browning-trail-cameras/
The firmware is available on my github site:
https://github.com/robertzak133/unified-btc-reverse
Note that my firmware does not work with later models HP5 (serial numbers starting 128 or greater), or with the new HP5-Ultra cameras. The latter has a new security feature which prevents this type of firmware hacking. So if you go down this path, you would need to find a source of earlier model cameras.
Alternatively, I am told that GardePro cameras offer an all-day/night timelapse mode right out of the box, but I have not tried this. In general, we find the image quality of the GP cameras to be lower than that of the Browning ReconForce/SpecOps models, but that may not be an issue for your project. In any case, you should check with vendor to make sure. See:
https://winterberrywildlife.ouroneacrefarm.com/2026/05/23/gardepro-a60-trail-camera-teardown-and-review/
You’ll have to experiment to find a timelapse frequency that will allow the batteries to cover your desired 72-hour target. For internal batteries, Li-Metal primary AA cells (e.g. Energizer Ultimate Lithium) are your best bet vs. other battery chemistries by something like a factor of two. On Browning cameras, operated through the night, you would probably need a frequency of 1 photo every 10 seconds, or perhaps every 30 seconds, to get through a continuous 72 hours on a set of 8 EUL AA cells. It will depend largely on duration of the night, due to the relatively higher energy required to operate the flashes. If you use an external battery, my firmware will take a photo every 1 second, max (factory firmware once after 5 seconds, max).
-bob
12 July 2026 5:03pm
Hi Bob,
Thanks so much for that incredibly detailed reply! That is extremely useful. I will look into all of those cameras. Thanks for sharing your solution for the Browning cameras.
Barbara
real‑time drone‑based telemetry tracking on forest‑dwelling bats in Europe
27 May 2026 3:33pm
8 July 2026 7:19pm
For reliable bat tracking in France and Western EU, look into SigFox enabled tags. For example these are today's smallest and lightest SigFox tags with Atlas Native support so you'll get a location in addition to basic telemetry: https://asd-tech.com/product/fx05-uwasp/
10 July 2026 5:54pm
Would appreciate knowing what you find out Garin, thanks.
12 July 2026 8:44am
Hi Garin,
If you're successful, please add it to the growing list of conservation drone projects on the Global Conservation Drone Map:
https://www.gctdf.org/map-conservation-drones
You can submit your project using the "Add Project" form here:
https://docs.google.com/forms/d/e/1FAIpQLSdTm3U87g4WV5L2GllGi8zjg-syJniLZA37sv0IWHUapj4e-w/viewform
It would be great to include it as part of this growing global resource.
DIY Noir Insect Camera Trap, solving the MIB paradox, looking for some advice.
9 June 2026 6:45pm
15 June 2026 9:57am
Depends a bit on how much activity you expect and how long insects stay around on average, but similar to what has been suggested before, I'd consider not using a motion trigger, but instead taking pictures in regular interval.
17 June 2026 1:06pm
Hello! The "MIB Paradox" is absolutely real, and PIR sensors are a dead end for insect detection due to their lack of thermal mass. However, falling back on a regular-interval time-lapse is a trap that will likely kill your battery and cost you the most interesting data.
Writing to an SD card is highly energy-intensive (causing current spikes often over 100mA). Doing this blindly every few minutes to save static pictures wastes a massive amount of power on empty frames. Worse, if a bug does show up, you only capture a single frozen moment and miss the entire temporal dynamic of its behavior.
Instead of a blind time-lapse, you can use your camera and a lightweight microcontroller (like an ESP32) as an Intelligent Optical Trigger. Here is how you can bypass the paradox:
This architecture gives you the best of both worlds. You get the energy efficiency of a motion trigger by treating disposable low-res frames as a "virtual PIR," and you only pay the heavy energy cost of lighting and SD card writing when there is guaranteed action to record.
12 July 2026 6:50am
Thanks Henri,
This sounds quite promising. So no Pi Zero at all, instead, I use Pico or ESP32. I am thinking about using the Seeed Studio XIAO ESP32S3 Sense with FOMO any thoughts on that?
Also, would it be an incredibly silly idea to just use one h264 webcam? Logitech?
I will follow your suggestion and let you know how it goes,Thanks a lot, bst
PantheraID: individual jaguar identification with computer vision, built from 14 years of camera trap data.
9 July 2026 5:39pm
11 July 2026 5:04pm
Would love to talk. We build the AI model MiewID (currently v4.1) and Wildbook, which has been deployed for jaguars on Whiskerbook.org. Happy to share ideas. Data cleaning and multispecies approaches to increase data volume and promote generalization are really the levers that have worked for us.
Wildlabs group - "Zoos and Aquariums"
3 November 2025 10:56pm
20 May 2026 2:45pm
This sounds like a great idea indeed, excellent!
I think the linking you suggest is a key role (perhaps the key role?) for zoos and aquariums, and so this group should advance that.
The "nexus effect" I mention at the top of this thread makes me think that a network/resource mapping project would be a great area of focus for this group. For example, to collaborate on tools to highlight exactly the types of linkages you suggest (maybe with something like kumu, graphistry, or Neo4J Bloom.)
Thanks very much for sharing this idea (and please share more!), it is something this group should work on and make happen.
10 July 2026 9:14am
Hi Stephen, Hi All!
I'm very keen to see this group be created and bring together a wide audience of those interested in Z&A's role in conservation tech. I have recently started as a researcher at Marwell Wildlife / the University of Surrey in the UK, and I like many of you I am particularly interested in how to transfer utility from Zoo research to in-situ contexts.
Looking forward to chatting to you all as this group grows!
11 July 2026 2:09pm
I am very excited about the group being launched and am eagerly awaiting it! This platform is such a unique hub for networking and support.
Research Coordinator II - Seasonal - San Diego Zoo Wildlife Alliance
11 July 2026 12:21am
RFID Smart traps
23 April 2021 9:10am
6 January 2023 8:56pm
Here is another RFID resource/vendor:
11 August 2023 12:50pm
Hi Caitlin
I'd recommend https://naturecounters.com/ who from past experience will work with you to come up with a good trap design. Their approach is to use an IR to detect when an animal starts to pass through the detector, which then triggers the RFID coil to be powered up. The huge advantage of this method is the battery power required is then very low, and the data can be stored into an SD card, all in one cheap, self-contained unit.
Roy
10 July 2026 10:58pm
Hi Sam, I'm looking to make this for my study species. I sent a message via the website but haven't hear anything back. Are you all still making these?
New "Human Dimensions" group on Wildlabs?
29 May 2026 7:25pm
26 June 2026 6:33pm
I would be interested
10 July 2026 3:43pm
Great idea!. I am interested in joining.
10 July 2026 5:42pm
Sounds great Matt!
GPS tags for medium-sized parrots
15 September 2025 4:40pm
30 June 2026 9:38pm
Have you received any feedback on this? I am also looking for similar advice for a crow species.
10 July 2026 2:58pm
Hey @cpespanola and @laurakb , I've seen @andreassenn at Copernicus Technologies works on different sensors for bird tracking without GSM, might be interesting for your work!
Mini AI Wildlife Monitor
25 June 2025 12:27pm
5 July 2026 5:19pm
Wow, what a great project.
8 July 2026 8:36pm
This is a great project! Some comments:
RaspberryPI though accessible is not the best fit for video pipelines and AI workloads or off grid deployments:
- it lacks onboard ISP which means either software implemented ISP, distorted data or on camera non energy optimized ISP.
- it lacks any power management techniques, low power modes, etc.
- it runs from SDCard using the same one for OS, swap and data, any corruption can lead to full loss.
- it runs any AI/ML workload on CPU which is extremely non efficient and any addon accelerators such as Hailo8 add a lot to power consumption and heat dissipation representing more challenges.
The advantages are of course plenty of documentation, community and all kind of makers addons, hats, etc.
For something more realistic, real life suitable I would suggest using something based on SoC with integrated NPU such as Hailo 15, Renesas RZ/V, Synaptics SL1680, MediaTek Genio or even the I.MX8M Plus for very light AI/ML workload. All of these have variety of SBCs, kits or even standalone smart camera oriented designs available from different vendors.
10 July 2026 1:32am
Yes, there are quite a few SBCs that use SoCs with integrated NN acceleration.
Except I think you are massively downplaying the advantages of the Raspberry Pi
"plenty of documentation, community and all kind of makers addons, hats, etc."
That is quite literally everything that makes the Raspberry Pi.
I've played with plenty of SBCs that are cheaper and have better specs than the Raspi, but they are almost useless because of the lack of "documentation, community and all kind of makers addons, hats, etc."
For a purpose built product by a team of engineers (with a lot of time and money behind them) then these SoCs with inbuilt NN are likely the future of this for of edge Ai deployment. But unless someone develops a well supported and well documented, general purpose device that uses one of these SoCs, then the default will still be the RasPi.
Raspberry Pi Alternatives for Edge AI!
6 July 2026 7:10am
8 July 2026 12:01pm
Hi Luke,
Really enjoyed another one of your videos, you've definitely got me experimenting more with edge compute and AI in my garden. I was wondering whether you’ve done much with NVIDIA Jetson boards? I think you mentioned having an older version in a previous video.
I got an Orin nano super developer kit since the higher-memory RPis doubling in the past year, while the Orin Nano/Orin Nano Super now seem surprisingly fairly static in price over that time period. What interests me is the possibility of using them for wildlife monitoring: running detection or ID models locally, collecting new field data, self labelling and training then using that data to periodically improve the model all on the edge.
I’m not sure whether full on-device training would be practical, but perhaps a more realistic workflow would be active learning or a teacher–student/pseudo-labelling setup: the device flags uncertain or interesting detections, those get reviewed or labelled, and the model is retrained or fine-tuned periodically self redeploying via scripts or a local model and agentic harness.
There also seems to be a good ecosystem of cameras and sensors around Jetson. I assume you find limitations in the Jetson boards for ecology/wildlife-monitoring projects?
10 July 2026 1:23am
I think you'd find a system that tried to self-train on the edge very complicated for not much benefit.
The drawback of edge deployment of NNs is the power requirement, you therefore want to have your edge node running as lean as possible.
Running self-supervised or semi-supervised learning on the edge adds huge overhead in the device requirements. Also if you have multiple devices you would want them all running the same model, to ensure consistency across sites.
You will likely need a human to review and label new data before training to prevent compounding model degradation, which mean a strong network connection, at which point transmitting data to the cloud for cloud based training would be much more efficient as you could collate the new data with a much larger dataset etc. However, in saying this, for most applications you don't really need to update the model once deployed, or if you do, it does not need to be done often (max - maybe once every few months or so?).
The Jetson boards you buy off-the-shelf are development kits, not really designed for "real-world" use cases, hence their somewhat odd form-factor (they're designed to sit on a desk not in an outside in an enclosure). You would ideally design your own carrier board for for the compute module with required peripherals etc.
People do use them for edge based ecology/wildlife monitoring as they are easier to get set-up with off-the-shelf models, but I think they are a bit overkill, especially since they use quite a bit more power than other options.
From Scratch Tutorial on AI Object Detection
10 July 2024 2:32pm
22 October 2025 5:25pm
We started getting frustrated with fiftyone, and are working with a friend to build really nice user friendly data labeler. We will share here when it is ready! Maybe that can help you out?
I got frustrated with 51 because there were some weird bugs in terms of the user interaction that really started holding us back
28 October 2025 9:17am
Hi Andrew,
thanks for your reply, good to know that fiftyone is not the perfect solution. Excited to see your data labeller when its finished :D
9 July 2026 8:56pm
Update: I am retraining with new Yolo26 and a bigger dataset we made. I made some improved scripts for
and
that are both available here: https://github.com/Digital-Naturalism-Laboratories/Mothbox/tree/main/AI/training_scripts
they have features like generating thumbnails to double check all your work before you train on it and disabling or tweaking features depending on your memory configurations and such! enjoy!
Aerospace Engineering student scoping for masters final year project- where are the monitoring bottlenecks?
9 July 2026 1:39pm
Conservation Genetics Post Doctoral Associate - San Diego Zoo Wildlife Alliance
8 July 2026 5:28pm
Post Doctoral Associate, Elephant Communication - San Diego Zoo Wildlife Alliance
8 July 2026 5:23pm
New paper calls for animal movement indicators in biodiversity policy – GEO BON
8 July 2026 1:48pm
New paper calls for animal movement indicators in biodiversity policy – GEO BON
Mothbox - Upcoming Features
7 March 2025 8:03pm
10 March 2025 3:45pm
@hikinghack that's a complete update!
A lot of compassion regarding the budget situation, especially knowing how great @briannaljohns is.
It’s great to see a clear plan for the hardware development! I hope you'll have nice outcomes from it!
29 June 2026 8:53pm
On the microcontroller approach for power gating the Pi — this is a well-proven pattern and worth doing even if the idle power is already low, because it also gives you a hardware watchdog for free.
The typical architecture is a small ultra-low-power MCU (STM32L0, SAMD21, or even an ATtiny) that stays awake in stop mode drawing a few microamps, handles the RTC wake-up schedule, and drives a P-channel MOSFET or load switch to cut power to the Pi entirely. When the Pi is done with its task, it signals the MCU via a GPIO line and the MCU cuts power and goes back to sleep. The key advantage over software sleep is that even a hung Pi gets cut off at the next scheduled cycle — no manual intervention needed in the field.
For the Mothbox use case where the Pi might sit idle for weeks, even 10mA idle current adds up to about 168mAh per week — enough to matter for a solar-charged system during a cloudy period. A properly gated system can get standby down to under 1mA including the MCU and RTC.
One practical note on the integrated PCB direction — if you're planning to integrate the MCU on the same board as the Pi interface, make sure to include a dedicated programming header (SWD for STM32, UPDI for modern ATtiny) so the MCU firmware can be updated in the field without disassembly. It's a small thing that saves a lot of frustration during deployment.
Happy to share more detail on the power switching circuit if useful.
7 July 2026 6:16pm
THanks for this!!!
Announcing the WILDLABS Awards 2026 Grantees
7 July 2026 2:27pm
7 July 2026 6:16pm
way to go! so many rad projects like always!