🔥 Popular | Latest

Bad, Funny, and Inception: FeaturesBusiness Explore Marketplace Pricing This repository Search Sign in or Sign up tensorflow/tensorflow O Watch5,517St62,04Fork 29,982 Code Issues 727 Pull requests 53 Projects Insights having problem to identity porn images -> especaly with penises (NO its not a joke!!) #11147 New issue Open puckpuck85 opened this issue 4 hours ago 1 comment puckpuck85 commented 4 hours ago Assignees No one assigned i m running a website where a mass on photos are uploaded (upload without registration)... so i get aaaaa lot of penis-trolls or whatever makes them post their private parts... anyhow Labels None yet i tried to train inception model with tensorflow (newest version 1.x) so i made a folder with "penises (approx 160 pics) and one with different images with person how dont show their penis (approx 160 pics.. Projects training accuracy is quit good-> over 90 %-but testing the model with other pictures-it fails really bad on None yet hmmmm, i know.. 160 pics are not that much for training, but i thing the problem is: No milestone guy on the beach in shorts is: ok guy on the beach with penis lurking out of his pants: is not okay but the difference is quite small between the pictures... because a penis is (mostly) quite a small part of the human body. so hard to detect... anybody could help? and no, its sounds funny, buts no joke.. i suffering under all the uploads i have to review manualy.... it's really not funny to see over 1000 penises a day : ( cheers, puck xdumaine commented 21 minutes ago This sounds like a really hard problem. Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment If not hotdog{return True}
Save