What causes "saída do modelo com formato inesperado: {:?}" in tonhowtf/omniget?

Verify the model_id maps to the correct segmentation ONNX file and that outputs[0] is the mask output

saída do modelo com formato inesperado

Error message

saída do modelo com formato inesperado: {:?}

What it means

A hand-written validation error: the model's output tensor has fewer than 2 dimensions, so the code cannot take the last two dims as mask height/width. This indicates the selected ONNX file is not the expected image-matting model (which should emit an NCHW or at least [.., H, W] mask), or its output was misidentified.

Solutions

Verify the model_id maps to the correct segmentation ONNX file and that outputs[0] is the mask output
Check the model's output shapes offline (netron or session.outputs) — expect rank 4 like [1,1,H,W]
If the model emits [N, C] logits, reshape/normalize before treating it as a spatial mask
Re-download the correct model if the file was replaced or partially overwritten

Example fix

// before
if shape.len() < 2 {
    return Err(anyhow!("saída do modelo com formato inesperado: {:?}", &shape[..]));
}
// after
if shape.len() < 4 || shape[1] != 1 {
    return Err(anyhow!("saída do modelo com formato inesperado: {:?} — esperado [N,1,H,W]", &shape[..]));
}

Defensive patterns

Strategy: validation

Validate before calling

let out_shape = session.outputs[0].output_type.tensor_shape();
if out_shape.rank() < 2 {
    return Err(anyhow!("modelo incompatível: saída tem rank {}, esperado >= 2", out_shape.rank()));
}

Type guard

fn is_spatial_mask(shape: &[i64]) -> bool {
    shape.len() >= 2 && shape[shape.len() - 1] > 0 && shape[shape.len() - 2] > 0
}

Try / catch

if !is_spatial_mask(&shape) {
    return Err(anyhow!("saída do modelo com formato inesperado: {:?} — verifique se o model_id aponta para um modelo de segmentação", &shape[..]));
}

Prevention

Validate model output rank/shape when registering a new model_id in params_for
Inspect new ONNX files with Netron before wiring them in
Fail fast with a startup smoke test that checks output shape on a dummy input

When it happens

Trigger: `shape.len() < 2` after extracting outputs[0] — e.g. a model that returns a scalar, a 1-D class-probability vector, or when the wrong output index was picked.

Common situations: Using a classification model instead of a segmentation model; a model export that collapsed/omitted spatial dimensions; accidentally pointing `session_for` at a different ONNX file (wrong model_id to file mapping).

Related errors

AI-assisted analysis of tonhowtf/omniget@8600b91f42 (2026-09-12). Data as JSON: /api/errors/b2b46361fefbd056. Report an issue: GitHub.

Appendix: source

Thrown at src-tauri/omniget-core/src/core/tools/img_bg.rs:387

    session: &mut ort::session::Session,
    img: &DynamicImage,
    p: &BgParams,
) -> anyhow::Result<GrayImage> {
    let side = p.size as i64;
    // O `ort` traz um `ndarray` próprio (0.17) e o crate usa o 0.16, então o
    // tensor atravessa a fronteira como forma + dados contíguos, que é o que o
    // `Tensor::from_array` aceita sem depender de versão de crate nenhuma.
    let (data, _) = normalize_input(img, p).into_raw_vec_and_offset();
    let tensor = ort::value::Tensor::from_array((vec![1, 3, side, side], data))
        .map_err(|e| anyhow!("não montei o tensor de entrada: {e}"))?;
    let outputs = session
        .run(ort::inputs![tensor])
        .map_err(|e| anyhow!("a inferência falhou: {e}"))?;
    let (shape, raw) = outputs[0]
        .try_extract_tensor::<f32>()
        .map_err(|e| anyhow!("não li a saída do modelo: {e}"))?;
    if shape.len() < 2 {
        return Err(anyhow!(
            "saída do modelo com formato inesperado: {:?}",
            &shape[..]
        ));
    }
    let h = shape[shape.len() - 2].max(0) as u32;
    let w = shape[shape.len() - 1].max(0) as u32;
    let small = mask_from_raw(raw, w, h)?;
    let (ow, oh) = (img.width(), img.height());
    Ok(image::imageops::resize(
        &small,
        ow,
        oh,
        FilterType::Lanczos3,
    ))
}

fn run_blocking(
    opts: &BgOptions,

View on GitHub (pinned to 8600b91f42)