---
title: Embroidery Complexity API
sdk: docker
app_port: 7860
---

# Embroidery Complexity API

`POST /predict` with a multipart field named `image`.

`GET /health` checks whether the model is available.

## Protected training endpoint

The Space also exposes a protected fine-tuning route. Configure these Space
Secrets before using it:

- `TRAINING_API_TOKEN`: a separate long random token used by the training client.
- `HF_TOKEN`: a Hugging Face token with write access to this Space.

The dataset ZIP must contain these folders, with at least six images in each
class and at least three different orders per class:

```text
baixo_risco/
medio_risco/
alto_risco/
```

Start a job:

```bash
curl -X POST "https://vitordigitizing-embroidery-complexity-api.hf.space/train" \
  -H "X-Training-Token: YOUR_TRAINING_API_TOKEN" \
  -F "dataset=@dataset.zip" \
  -F "epochs=20"
```

The response contains a `job_id`. Poll `GET /train/{job_id}`. When the job is
completed, the best `complexidade_best.pth` checkpoint is committed to this
Space and the running API is updated in memory. The route is intentionally not
open without `TRAINING_API_TOKEN`.