Image rating (safe, r15, r18) prediction model.
Please note that due to the often unclear boundaries between safe, r15 and r18 levels, there is no objective ground truth for this task, and the data is scraped without any manual filtering. Therefore, the models trained on this dataset can only provide rough judgments. If you require an accurate solution for classifying R18 images, it is recommended to consider a solution based on keypoint object detection.
| Model | FLOPs | Accuracy | Confusion Matrix | Description |
|---|---|---|---|---|
| caformer_s36_plus | 22.10G | 74.26% | Confusion Matrix | Model: caformer_s36.sail_in22k_ft_in1k_384 pratrained from timm |
| mobilenetv3 | 0.63G | 64.77% | Confusion Matrix | Model: mobilenetv3_large_100 from timm |
| mobilenetv3_sce | 0.63G | 66.27% | Confusion Matrix | Model: mobilenetv3_large_100 from timm, use SCELoss as loss function |
| mobilenetv3_sce_dist | 0.63G | 69.49% | Confusion Matrix | Distillated from caformer_s36_plus, using mobilenetv3_large_100 |