lh-nsfw-dp-detector

RT-DETRv4-L (HGNetv2-B4) — NSFW Detection

Real-time object detection model based on RT-DETRv4 with HGNetv2-B4 backbone, trained on a custom 2-class NSFW detection dataset.

Model Details

Usage

Installation

git clone https://huggingface.co/Bjoon/lh-nsfw-dp-detector
cd lh-nsfw-dp-detector
pip install -r requirements.txt

Inference

import torch
import torch.nn as nn
import torchvision.transforms as T
from PIL import Image
from engine.core import YAMLConfig

# Load config and checkpoint
cfg = YAMLConfig("configs/rtv4/rtv4_hgnetv2_l_coco.yml", resume="best_stg1.pth")
cfg.yaml_cfg['HGNetv2']['pretrained'] = False

checkpoint = torch.load("best_stg1.pth", map_location="cpu")
state = checkpoint['ema']['module'] if 'ema' in checkpoint else checkpoint['model']
cfg.model.load_state_dict(state)

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.model = cfg.model.deploy()
        self.postprocessor = cfg.postprocessor.deploy()
    def forward(self, images, orig_target_sizes):
        outputs = self.model(images)
        return self.postprocessor(outputs, orig_target_sizes)

model = Model().to("cuda").eval()

# Run inference
im = Image.open("test.jpg").convert("RGB")
w, h = im.size
transforms = T.Compose([
    T.Resize((640, 640)),
    T.ToTensor(),
    T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])
im_data = transforms(im).unsqueeze(0).to("cuda")
orig_size = torch.tensor([[w, h]]).to("cuda")

labels, boxes, scores = model(im_data, orig_size)

CLI Inference

python tools/inference/torch_inf.py \
  -c configs/rtv4/rtv4_hgnetv2_l_coco.yml \
  -r best_stg1.pth \
  --input image.jpg \
  --device cuda:0

File Structure

├── best_stg1.pth                          # Model checkpoint
├── configs/                               # YAML config chain
│   ├── rtv4/rtv4_hgnetv2_l_coco.yml
│   ├── dfine/dfine_hgnetv2_l_coco.yml
│   ├── base/{dfine_hgnetv2,optimizer,dataloader,rtv4}.yml
│   ├── dataset/{custom_detection,coco_detection}.yml
│   └── runtime.yml
├── engine/                                # Model + inference code
├── tools/inference/torch_inf.py           # Inference script
└── requirements.txt

License

CC BY-NC-ND 4.0 (Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International)

This model is provided for non-commercial research and personal use only. Commercial use of any kind is strictly prohibited, including but not limited to:

• Integration into commercial products or services
• Use in revenue-generating applications
• Deployment in business operations

For any questions regarding licensing, please contact the author.