Naven
Glossary
Image screening
Checking a picture against a policy before it is delivered.
Harder than text and more necessary. Classifiers handle obvious categories well and context badly — a medical photograph, a classical painting and a news image all read as the thing they depict. False positives on images are more annoying than on text because the sender usually cannot tell what was objected to.
Running two independent checks and requiring both to pass trades a higher false-positive rate for a much lower miss rate.
Where Naven stands
Every image uploaded to Naven is checked by two independent models and has to pass both.