buystufflive/safety-classifier

Live Safety Classifier

Live Safety Classifier predicts the content-safety category of one image: hot, nsfw, or safe.

The model is fine-tuned from timm/mobilenetv4_conv_medium.e250_r384_in12k.

What It Is For

This model is built for filtering images. An application can use it to find which safety category an image belongs to before running more expensive analysis.

Files

File	Purpose
onnx/model.onnx	ONNX model for CPU/server inference.
onnx/model.fp16.onnx	Optional FP16 ONNX candidate.
onnx/model.onnx.data	External ONNX weight data loaded beside model.onnx.
model.safetensors	PyTorch state dict for reproducibility and continued training.
config.json	Model identity, base model, output names, and safety labels.
preprocess.json	Resize and normalization contract used during training/export.
train.json	Sanitized training recipe for post-training or reproduction.
inference/python.py	Python helper for ONNX inference from image paths.
inference/typescript.ts	TypeScript helper for ONNX inference from image paths.
inference/labels.json	Safety labels used to decode model logits.
README.md	This model card and runnable inference reference.

Outputs

Read the safety output:

Output	Shape	Meaning
safety	[batch, safety_class_count]	Content-safety logits.

Inference uses argmax in this version.

The model also has a phone classification head, which outputs screen. It is not used: only the safety head is.

Test Results

Model	Test	Accuracy	Balanced Acc	Macro F1	Top-2	Images
fp32	full test	0.9575	0.8941	0.8600	0.9921	23615
fp32	safety-balanced test	0.8957	0.8957	0.8947	0.9847	3000
fp16	full test	0.9575	0.8941	0.8600	0.9921	23615
fp16	safety-balanced test	0.8957	0.8957	0.8947	0.9847	3000

CPU Timing

Model	Images/s	Load Mean	Model Mean	Total Mean	Total Median	Total P95	Provider
fp32	20.0538	0.12 ms	49.52 ms	49.87 ms	57.84 ms	85.82 ms	onnxruntime:CPUExecutionProvider
fp16	20.8351	0.11 ms	47.69 ms	48.00 ms	48.60 ms	63.65 ms	onnxruntime:CPUExecutionProvider

Timing is measured with ONNX Runtime CPU execution on Apple M4 Max (16 logical cores). Total latency includes image load/preprocess, model inference, and label decoding.

Classification Labels

For the safety label list, see config.json.

Preprocessing

Use the preprocessing contract in preprocess.json.

1. Read the image.
2. Apply EXIF orientation.
3. Convert to RGB.
4. Resize so the longest side is 1024 pixels and keep the original aspect ratio.
5. Do not crop, stretch, square-pad before preprocessing, or horizontally flip.
6. Convert to float32 in [0, 1].
7. Normalize with the mean and standard deviation from preprocess.json.
8. Pad each normalized tensor to the batch maximum height and width, rounded up to a multiple of 32.

The exported model accepts dynamic batch, height, and width.

ONNX Inference

The helpers return only safety.

Keep onnx/model.onnx.data beside onnx/model.onnx; ONNX Runtime loads the external tensor data when it opens the graph. By default the helpers load the FP32 model at onnx/model.onnx. Pass fp16 to load onnx/model.fp16.onnx.

Download

Download the exported model folder from Hugging Face:

from huggingface_hub import snapshot_download

model_dir = snapshot_download("buystufflive/safety-classifier")

Python

Install the runtime dependencies:

python -m pip install numpy pillow onnxruntime

Import the helper from the exported model folder:

from inference.python import Classifier, classify

print(classify("example.png"))
print(classify("example.png", model_format="fp16"))

classifier = Classifier(model_dir)
print(classifier.classify_batch(["one.png", "two.png"]))

fp16_classifier = Classifier(model_dir, model_format="fp16")
print(fp16_classifier.classify_batch(["one.png", "two.png"]))

Or run it directly:

python inference/python.py example.png another.png --model-format fp16

TypeScript

Install the runtime dependencies:

bun add sharp onnxruntime-node

Import the helper from the exported model folder:

import { Classifier, classify } from './inference/typescript.ts';

console.log(await classify('example.png'));
console.log(await classify('example.png', modelDir, 'fp16'));

const classifier = await Classifier.create(modelDir);
console.log(await classifier.classifyBatch(['one.png', 'two.png']));

const fp16Classifier = await Classifier.create(modelDir, 'fp16');
console.log(await fp16Classifier.classifyBatch(['one.png', 'two.png']));

Or run it directly:

bun inference/typescript.ts example.png another.png --model-format fp16

Citation

If you use this model, please cite:

@misc{safety-classifier,
  title={Live Safety Classifier},
  year={2026},
  publisher={Live},
  url={https://huggingface.co/buystufflive/safety-classifier}
}