TimTheJedi There are different stages in the recognition. The one you have issues with is face detection. This one is not trainble. It does not learn from your input. Most of the time it works good but sometimes it gives false positives. We can set the threshold higher, it then will give less false positives, but also less true positives too. So this is tradeoff.
As to faces suggestion - this one gets trained on your input if you reject faces. However, when you ignore face it gets completely deleted from the database and does not take part in any training. But there is no point to train on sofas anyway, so I would suggest to conitnue deleting them.
