likewendy/yolo26n-cls-porn

yolo26n cls porn space

Online demo for the yolo26n cls porn model.

yolo26n cls porn is an adult content image classification model trained on YOLO26n-cls, which classifies images or video frames into 35 adult content categories.

Features

• —Image labeling: upload an image to get the top-5 predicted classes sorted by confidence; the label accepted by the per-class confidence threshold and top1/top2 margin is highlighted.
• —Video labeling (<=30MB): frames are sampled about every 10 seconds with OpenCV; frame labels are filtered by the repetition counting rules from repetition_counting_rules.json.

Inference runs on CPU (the model has only ~3M parameters, so CPU inference takes about 0.5s per image — much faster than Zero GPU, which adds queueing and CUDA init overhead). The model is loaded at startup from the gated model repository likewendy/yolo26n-cls-porn.

The Space runs on Zero GPU hardware (free tier), but inference itself uses only the CPU: the @spaces.GPU decorated _zero_gpu_probe function exists solely to pass Zero GPU's startup check ("No @spaces.GPU function detected during startup") and is never called, so requests are not queued for a GPU.

Usage

pip install -r requirements.txt
python app.py