the 4th person in the image below is consistently missed, although the model is usually very good at detection even with back of head etc. Any suggestions on why they are missed? Note I am running in
And a shot from above, same person missed
Hello @robmarkcole This is an interesting case. You can try lowering the min_confidence to see how it responds. We are planning to add support for detecting specific objects soon. This would lead to better accuracy. Towards this end, we have released a new version of our open source tool, ImageAI to allow training of custom objection detection models.
Future releases of DeepStack will add support for custom object detection models trained with ImageAI.
As a follow up to the above, you can learn more about using the new ImageAI release via the tutorials linked below.
@OlafenwaMoses something strange, the guy that previously was hard to detect today has TWO bounding boxes, any ideas whats going on?
Hello @robmarkcole , are you running in the same mode? Also, did you explicitly specify a minimum confidence?
Hi @john, running in HIGH mode. Same issue, different person
docker run -v localstorage:/datastore --name="deepstack"\
-e VISION-DETECTION=True \
-e VISION-FACE=True \
-e MODE=High \
-p 5000:5000 -d \
Hello, in this case, the min_confidence is too low. Set it to at least 0.5. As you can see, the second bounding box had a confidence of 0.491.