---
license: other
license_name: nutrient-commercial
pipeline_tag: object-detection
language: [en]
tags:
  - object-detection
  - document-ai
  - form-field-detection
datasets:
  - nutrientdocs/form-field-v1-benchmark
metrics:
  - accuracy
---

# form-field-v1-medium · _commercial_

**The highest-accuracy tier of the `form-field-v1` detector family** — locates `Text`, `Choice`
(checkbox/radio), and `Signature` widgets on **empty, filled, and handwritten** document pages. ~34M params,
NMS-free, single-resolution 1216.

The runnable weights are commercial and are not downloadable from this repository — this public page is the
product specification and scorecard. Prefer a small, open model? See
[`form-field-v1-nano`](https://huggingface.co/nutrientdocs/form-field-v1-nano) (Apache-2.0).

- 🎯 **Try it:** [form-field-v1-demo](https://huggingface.co/spaces/nutrientdocs/form-field-v1-demo)
- 🏆 **Leaderboard:** [form-field-v1-leaderboard](https://huggingface.co/spaces/nutrientdocs/form-field-v1-leaderboard)
- 📊 **Benchmark:** [form-field-v1-benchmark](https://huggingface.co/datasets/nutrientdocs/form-field-v1-benchmark)

## Results

COCO **mAP50-95** on the [`form-field-v1-benchmark`](https://huggingface.co/datasets/nutrientdocs/form-field-v1-benchmark),
one shared pycocotools scorer across all models.

| Model | Empty | Filled | Handwritten |
|---|---|---|---|
| **form-field-v1-medium** (this model) | **0.529** | **0.783** | **0.764** |
| FFDetr (baseline) | 0.461 | 0.170 | 0.225 |
| FFDNet-L (baseline) | 0.373 | 0.285 | 0.285 |

Per-class on empty (mAP50-95): Text 0.605 / Choice 0.614 / Signature 0.369. On the neutral CommonForms-test it
reaches 0.361. Full per-class breakdowns are on the
[leaderboard](https://huggingface.co/spaces/nutrientdocs/form-field-v1-leaderboard).

## Output

```json
[ { "box": [82, 164, 336, 41], "type": "Text", "score": 0.94 } ]
```

`box` is `[x, y, w, h]` in page pixels; `type` ∈ `Text`, `ChoiceButton`, `Signature`. Served as a single-file
ONNX graph (fp32 + fp16), verified to reproduce the benchmark (fp32 mAP 0.6909 / fp16 0.6860 over all variants).

## Intended use and limits

- Empty, filled, and handwritten digital/scanned forms are all supported.
- GPU/server model (~57 fps on an A100). For on-device CPU use `form-field-v1-nano` / `-small`.
- Very dense pages benefit from higher input resolution.

## License & data

Weights are offered under a commercial Nutrient license. Evaluation is public and reproducible via
[`form-field-v1-benchmark`](https://huggingface.co/datasets/nutrientdocs/form-field-v1-benchmark). The FFDetr
baseline is [FF-DETR / CommonForms](https://huggingface.co/jbarrow/FFDetr) by jbarrow, Apache-2.0.

> ### 📩 Get access
>
> `form-field-v1-medium` is commercial and its weights are not downloadable here. To run it on-prem, with
> documents staying in your infrastructure — **contact Nutrient:
> [nutrient.io/contact-sales](https://www.nutrient.io/contact-sales/).**

## About the author

<a href="https://nutrient.io/">
  <img src="https://avatars2.githubusercontent.com/u/1527679?v=3&s=200" height="80" />
</a>

This project is maintained and funded by [Nutrient](https://nutrient.io/) - The deterministic document infrastructure enterprises run their highest-stakes workflows on: replayable output, clear exceptions, and full audit trails on the messy, regulated documents where AI alone breaks.