Hi Dan,

The team that need to use this have limited tech/coding skills and time to learn so we're trying to find a solution that isn't too complicated. Not to say that MegaDetector is but I just think in this scenario it might be a bit too much. The team are just interested in identifying birds flying from nests and people. In this case the simpler the better just wondering if there is anything in the middle ground between using photo editing software and running lots of scripts :)

Best,

Chelsea

Got it.  Sounds like it's not MegaDetector that you want to avoid per se, it's the command line, or the R interface that WildCo-FaceBlur provides.

I don't know of anything that does face-blurring off-the-shelf that (a) runs locally (there are cloud services that do this that aren't related to camera traps) and (b) doesn't require some command-line stuff, but this is a pretty common need, so it would be good to come up with something that checks the "no-command-line" and "no-code" boxes.

An ideal solution might be to add a face-blurring step to AddaxAI (formerly EcoAssist), which is a GUI that wraps up MegaDetector (and other models, but MegaDetector is the one that is relevant to this discussion).  No code required.  AddaxAI does other postprocessing steps like this, just not this specific thing (face-blurring) yet.  Maybe a nice open-source contribution someone could make to AddaxAI, since this is a pretty self-contained change. @pvlun 

I'll buy a fancy coffee for anyone who adds this to AddaxAI (but you have to come to Seattle to claim that fancy coffee).

I think this pretty much sums up what we need! Would be great to see this feature added to AddaxAI, or know if anyone is working on something similar 

@dmorris  I'll come and collect that fancy coffee one day!

@Chelsea_Smith I've added the functionality to blur MD detections of people to the latest AddaxAI release. Note that it only works on images. Visualising videos is something else... Also, I've checked it on my computer and think it should work without any problems, but I haven't had the chance to thoroughly test it. Please go ahead and try some different settings to find the edge cases!

Read "AddaxAI - Simplifying camera trap image analysis with AI" on Addax Data Science

AddaxAI - Simplifying camera trap image analysis with AI

Previously known as EcoAssist. An AI platform to analyse images with machine learning models for automatic detection, so that ecologists can save time and focus on conservation efforts.

Addax Data Science

(Image taken form iStock) 

Well look at that, <48 hours from feature request to release!  Yes, Peter, I owe you coffee.  This was so fast that I'll buy you two coffees.

I was also inspired by this thread to add this to the MD Python package too, so this is now present (pending a merge into main and an update to PyPI) in separate_detections_into_folders.py and visualize_detector_output.py.  Specifically, both now take an optional list of category names to blur; one would typically specify ['person'] or ['person','vehicle'].  But most users should ignore this reply and just use AddaxAI for this. :)

That was indeed impressively fast! I thought about having a go at it myself but a direct integration with AddaxAI or directly in MD makes so much sense! Well done! 

Now how about a classifier to work with MegaDetector leveraging the 

imageomics/bioclip-vit-b-16-inat-only model ...?

Read "imageomics/bioclip-vit-b-16-inat-only · Hugging Face" on

imageomics/bioclip-vit-b-16-inat-only · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

I believe it could be a great help for multi-species annotation tasks.

I made a working hack using the https://huggingface.co/imageomics/bioclip-vit-b-16-inat-only

Read "imageomics/bioclip-vit-b-16-inat-only · Hugging Face" on

imageomics/bioclip-vit-b-16-inat-only · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

 to classifiy MegaDetector detections for my multispecies images. 

I am doing this via Fiftyone

@pvlun Amazing that was a quick turn around - this is going to be so useful in our work thank you so much!

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!

Thank you for sharing!