Yes. I think this is really the way to go!

Here is another metadata initiative to be aware of. OGC has been developing a standard for describing training datasets for AI/ML image recognition and labeling. The review phase is over and it will become a new standard in the next few weeks. We should consider its adoption when we develop our own training image collections.

OGC Seeks Public Comment on New Standard for Training Data for AI/ML Applications - Open Geospatial Consortium

OGC TrainingDML-AI Standard defines the model and encodings for standardizing any training data used to train, validate, and test Machine Learning models that involve location or time.

Open Geospatial Consortium

Hi Michelle,

it would be good to get some more information on your project. What question are you trying to answer? How many cameras did you have deployed? How many different study sites/haul-out areas are there?

For the more computationally-inclined, the Project Zamba software can also be used as a python package and the project codebase is fully available on GitHub.

@hikinghack @Nycticebus-scientia We just ran an open-source competition on this exact question! DrivenData is a small group of data scientists, and we partnered with the Max Planck Institute for Evolutionary Anthropology (MPI-EVA) and the Wild Chimpanzee Foundation (WCF) to compile enough hand-labeled data to apply machine learning to the depth problem. In your original post, this falls under the AI Prediction Based category. The code behind all of the top-performing models is freely available on GitHub. We're hoping to make these freely available in the future in the a more user-friendly way.

@JHughes to your point about motion sensor camera traps being continuously set off -our free, open-source web application called Zamba Cloud uses machine learning to automatically sort videos by either what animal they contain or whether they are blank. You just upload a set of videos, and the web application will output a list of labels for each video! This may not address the battery drainage problem, but can help with filtering out blank videos and identifying which actually captured animals.

(apologies for joining this thread late)  Has anyone tried using the LIDAR in one of the newer iPhones so-equipped as a trail cam trigger? (e.g. iPhone12 Pro), perhaps in conjunction with a nueral network that processes the depth map looking for animal triggers?  (I know this is liable to be a power-hungry trigger, for now, just looking at a proof of concept)  

Hi David

It appears that you have been looking for existing models, however, most existing models are trained on either COCO or some other very generic dataset. So, if you want to identify just animals, you may be better off training your own model. It seems no one in this thread mentioned yet that it is possible to do transfer learning on existing models, which keeps most of the "visual part" of the model as is, but just changes the classification part so it can identify other things. This way you can take an existing model trained on COCO and in a fraction of the time it takes to train a full model, just retrain that for your animals.

Also have a look at your requirements for the inferencing stage. Some models take long in training but are superfast in inference and others are slow in both cases but very accurate, etc. If you want semi-realtime inferencing, you are probably looking at single shot detectors (SSD), and not RCNNs.