| import gradio as gr | |
| import os | |
| DESC = """# Classify a hardcore video is anal or not | |
| WARNING! Leave now if you are less than 18 years old! | |
| * The input video must be a hardcore one, otherwise the prediction score is meaningless(for example, a blowjob video), to determine a video is hardcore or not you can use the [model](https://huggingface.co/spaces/porntech/sex-position-video) of mine. | |
| * Total 2 classes are supported: ["Non-Anal", "Anal"]. | |
| * The same as [model](https://huggingface.co/spaces/porntech/sex-position-video), input must be a mp4 file with less than 10 seconds, prediction for long videos is undefined. | |
| * Classification of a hardcore video being anal or variginal can be somehow quite difficult depending on the sex position and camera view because anal and vargina are quite close to each other. For example, it is hard to find a hardcore video is anal or varginal if the camera is the front view of doggystyle, but will be quite easy if the camera is the front view of reverse cowgirl. So the model is trying to decide how evident the video is anal, often the model can not predict a video as anal not because the video is varginal but due to the camera view. | |
| * AUC is ~0.97 on a validation set of 1200 video clips with anal as positive class, the sample size of anal vs non-anal is about 400:800. Considering this sample size, a score above 0.4 might indicate anal. | |
| I will soon be on job market and is now looking for full time or part time jobs focusing on developping AI models for sexy/NSFW videos/images, if you are interested in me or this work, feel free to contact porntech@126.com""" | |
| with gr.Blocks() as demo: | |
| with gr.Row(): | |
| gr.Markdown(DESC) | |
| with gr.Tab("anal classification for video"): | |
| gr.load("porntech/anal-internal", src="spaces", hf_token=os.environ["PORNTECH_SECRET"]) | |
| demo.launch() |