Ask HN: Recommendations for automated image content filter services?
I'm working on a side project which accepts user-generated content and I'm looking for a straightforward web service to which I can submit an image and have it classified against a dataset looking for content that might get me in hot water with the App Store, Play Store, or other bodies which might want to pull the plug on my service such as my hosting provider. We already have quite an extensive manual moderation system that can react to user reports, but I'd like to cut down the moderation workload to an easier level using an automated service. Ideally this service would have fine-grained classification (ie swimwear/underwear is fine, nudity isn't; a picture of an explosion is fine, images of weapons aren't) but I'd sacrifice some fine-grained control in order to keep costs down if necessary.
My aim is to have three possible outcomes: if the image has a strong confidence of being inappropriate the submission simply gets blackholed, if there's a less strong but still fairly high chance of the image being inappropriate it goes though but is added into the moderation queue, and if it's not suspected to be inappropriate nothing happens and the post is let through unimpeded.
This doesn't have to be perfect, nor is it the only (or even the primary) means of moderation for my use case. However, this does seem a good way of reducing the moderation workload and reducing the risk of some adversary submitting inappropriate content then complaining to Apple/Google/your server host etc.
Cheers in advance.