tonhowtf/omniget · error

não montei o tensor de entrada: {e}

Error message

não montei o tensor de entrada: {e}

What it means

This error wraps a failure from `ort::value::Tensor::from_array` when building the NCHW input tensor (shape [1,3,side,side]) for the background-removal ONNX model. The `ort` crate rejects the shape/data combination when the element count does not exactly match the declared shape, the data is not contiguous/aligned, or the type does not match the declared tensor type. It is thrown eagerly before any inference runs.

Solutions

1. Verify `normalize_input` resizes to exactly p.size x p.size and produces 3 RGB channels as f32, so data.len() == 3*side*side
2. Log data.len() and the expected 3*side*side next to the error to confirm the mismatch
3. Check that p.size matches the ONNX model's declared input dimensions (inspect with `onnxruntime` tools or session.inputs)
4. Keep the shape Vec and data produced from the same `p` value, not from stale/other params

Example fix

// before
let (data, _) = normalize_input(img, p).into_raw_vec_and_offset();
let tensor = ort::value::Tensor::from_array((vec![1, 3, side, side], data))?;
// after
let input = normalize_input(img, p);
assert_eq!(input.len(), (3 * side * side) as usize, "input buffer/shape mismatch");
let (data, _) = input.into_raw_vec_and_offset();
let tensor = ort::value::Tensor::from_array((vec![1, 3, side, side], data))?;

let expected = 3 * p.size * p.size;
let input = normalize_input(img, p);
if input.len() != expected {
    return Err(anyhow!("input buffer {} != expected {} (1x3x{}x{})", input.len(), expected, p.size, p.size));
}

fn is_valid_input(buf: &[f32], side: usize) -> bool {
    buf.len() == 3 * side * side && buf.iter().all(|v| v.is_finite())
}

When it happens

Trigger: Calling `Tensor::from_array((vec![1, 3, side, side], data))` where `normalize_input` produced a buffer whose length != 3*side*side, or `p.size` disagrees with the actual resized image dimensions, or the f32 Vec was mis-shaped.

Common situations: A model config whose `size` parameter was changed without updating `normalize_input`; switching to a model with a different input resolution; a refactor of `normalize_input` returning a different channel order or count; passing an RGBA (4-channel) buffer instead of RGB (3-channel).

Understand the failure class

Background: Tensor shape mismatch errors ("must have shape", "expected shape ... got ..."): when tensor dimensions disagree with what an op or layer was told to expect — this error's family across 6 libraries.

Related errors

• não criei o builder de sessão ONNX: {e}
• não carreguei o modelo {}: {}
• o ONNX Runtime {} ainda não está instalado. Instale pela tel
• não existe build oficial do ONNX Runtime para este sistema.
• não consegui carregar {}: {}

AI-assisted analysis of tonhowtf/omniget@8600b91f42 (2026-09-12). Data as JSON: /api/errors/26be7d2b5258776a. Report an issue: GitHub.

Appendix: source

Thrown at src-tauri/omniget-core/src/core/tools/img_bg.rs:379

    Ok(std::fs::metadata(dest).map(|m| m.len()).unwrap_or(0))
}

// ── Execução ───────────────────────────────────────────────────────────

/// Roda a inferência num arquivo já aberto e devolve a máscara no tamanho
/// original da imagem.
fn mask_for_image(
    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,

View on GitHub (pinned to 8600b91f42)