---
license: mit
tags:
- iron
- amd-npu
- ryzen-ai
- xdna
- image-cropping
- npu2
---

# GAIC-VGG16 crop scoring on the AMD NPU

An [IRON](https://github.com/amd/IRON) export of [https://github.com/bo-zhang-cs/GAIC-Pytorch](https://github.com/bo-zhang-cs/GAIC-Pytorch) for AMD Ryzen AI NPUs: the compiled NPU kernels (`.xclbin` + instruction streams) and the packed weights the [`taconite-gaic`](https://crates.io/crates/taconite-gaic) Rust runtime replays, over XRT or directly over the `amdxdna` driver.

The kernels are compiled for **NPU2 (AIE2P: Strix Point, Strix Halo, Krackan)** and will not load on NPU1 (Phoenix, Hawk Point).

## Download

```bash
hf download brishen/iron-gaic-vgg16-npu2 --local-dir gaic
# or, from an IRON checkout:
python scripts/hf_models.py download gaic --repo brishen/iron-gaic-vgg16-npu2 --out gaic
```

## Usage

```bash
cargo install taconite-gaic             # over XRT
# or, with no XRT at all (straight to the amdxdna driver):
cargo install taconite-gaic --no-default-features --features cli,direct
gaic gaic check                        # against the bundle's reference image
gaic gaic crop --out crops/ photo.jpg  # best crop overall and at 1:1, 4:3, 16:9
```

See [`taconite-gaic`](https://crates.io/crates/taconite-gaic) ([docs](https://docs.rs/taconite-gaic), [source](https://github.com/Brishen/taconite)) for the full API. The bundle is exported by IRON's `iron/applications/gaic`.

## Provenance

- Upstream model: [https://github.com/bo-zhang-cs/GAIC-Pytorch](https://github.com/bo-zhang-cs/GAIC-Pytorch) (license: `mit`; its terms apply to these weights)
- IRON commit: `be4bb37`
- Uploaded: 2026-09-23
- Files: 21, 29.2 MB