Hi! First thanks for creating this amazing software, I’m loving it.
I have to analyze like 3 tb of trail camera videos so it will take a long time, it would be ideal to just let it running (not all files of course but smaller chunks) in the night and when its done, it saves the usual files (identification, spreedshet and others) and then turn off the PC. Is it possible? If not, then this is a suggestion, it would be so useful for working for large amount of data.
Thanks again!
Hi @Bernardo,
Thanks for the kind words, happy to hear you like it.
Good news: running overnight in chunks already works. In projects mode you can queue as many folders as you want and they’ll be processed one after another. And based on your suggestion, the next version v7.0.9 will keep your computer awake while it’s working, so it won’t fall asleep halfway through the night. When the run is done, it lets go and your PC will sleep on its own schedule. All your results are safely stored at that point, so exporting the files the next morning only takes a minute.
Auto-shutdown I’m leaving out for now, since it would turn off the whole machine and anything else running on it. But with the sleep fix, an overnight run should just work.
Good luck with the 3 TB!
Cheers,
Peter
Thanks @petervanlunteren. I detected something about this method. I’m using SpeciesNet 4.0.2a, using a subset of just 80 species (less that the ones assigned to my country). If I run the model in “Analyse a folder mode” it works ok, but if I use it on project mode apparently it uses more species that in folder mode and that makes difficult to compare bolders using different methods. I notice this by using timelapse, importing the recogition file and then using the “select” mode to filter by species.
Good catch! but it is not what it looks like. The two modes use exactly the same species selection. The difference is the file you have imported.
A project does not make a recognition file for Timelapse, only “Analyse a folder” does that. So the JSON you found for the project run is the internal results.json inside the hidden .addaxai folder. That file is the raw model output: it keeps the top 5 guesses of SpeciesNet for every detection, before your species selection, the geofence and the smoothing are applied. That is why Timelapse shows many more species there. AddaxAI keeps it raw on purpose, so it can redo the postprocessing when you change a setting without running the AI again. It is not meant to be used as a result. It’s internal for AddaxAI.
The recognition file from the folder run is built from the final results, one label per detection, so that one is correct.
If you want to compare in Timelapse, run both folders with “Analyse a folder”. If you want to stay in project mode, use the CSV or Excel export from the Export page, those have the same filtered labels.
Hope that makes sense!
Cheers,
Peter