Yalantis/uCrop · error · CImgArgumentException

blur_anisotropic(): Invalid specified diffusion tensor field

Error message

blur_anisotropic(): Invalid specified diffusion tensor field (%u,%u,%u,%u,%p).

What it means

CImg's blur_anisotropic() expects the diffusion tensor field G to share the instance's width/height/depth and to have exactly 3 (2D tensor: P11,P12,P22) or 6 (3D tensor) channels. Any other shape throws this error before blurring starts.

Source

Thrown at ucrop/src/main/jni/CImg.h:44202

    /**
       \param G Field of square roots of diffusion tensors/vectors used to drive the smoothing.
       \param amplitude Amplitude of the smoothing.
       \param dl Spatial discretization.
       \param da Angular discretization.
       \param gauss_prec Precision of the diffusion process.
       \param interpolation_type Interpolation scheme.
         Can be <tt>{ 0=nearest-neighbor | 1=linear | 2=Runge-Kutta }</tt>.
       \param is_fast_approx Tells if a fast approximation of the gaussian function is used or not.
    **/
    template<typename t>
    CImg<T>& blur_anisotropic(const CImg<t>& G,
                              const float amplitude=60, const float dl=0.8f, const float da=30,
                              const float gauss_prec=2, const unsigned int interpolation_type=0,
                              const bool is_fast_approx=1) {

      // Check arguments and init variables.
      if (!is_sameXYZ(G) || (G._spectrum!=3 && G._spectrum!=6))
        throw CImgArgumentException(_cimg_instance
                                    "blur_anisotropic(): Invalid specified diffusion tensor field (%u,%u,%u,%u,%p).",
                                    cimg_instance,
                                    G._width,G._height,G._depth,G._spectrum,G._data);
      if (is_empty() || dl<0) return *this;
      const float namplitude = amplitude>=0?amplitude:-amplitude*cimg::max(_width,_height,_depth)/100;
      unsigned int iamplitude = cimg::round(namplitude);
      const bool is_3d = (G._spectrum==6);
      T val_min, val_max = max_min(val_min);
      _cimg_abort_init_openmp;
      cimg_abort_init;

      if (da<=0) { // Iterated oriented Laplacians
        CImg<Tfloat> velocity(_width,_height,_depth,_spectrum);
        for (unsigned int iteration = 0; iteration<iamplitude; ++iteration) {
          Tfloat *ptrd = velocity._data, veloc_max = 0;
          if (is_3d) // 3D version
            cimg_forC(*this,c) {
              cimg_abort_test;

View on GitHub (pinned to f788b534b4)

Solutions

Ensure G has the same XYZ dimensions as the image (use resize or compute from the same source)
Ensure G._spectrum is exactly 3 for 2D tensors or 6 for 3D tensors
Add img.is_sameXYZ(G) && (G.spectrum()==3 || G.spectrum()==6) as an assert before calling

Example fix

// before
CImg<float> G = img.get_gradient().resize_halfXY();
img.blur_anisotropic(G);
// after
CImg<float> G = img.get_gradient(); // same XYZ, spectrum reshaped as needed
G.resize(img.width(), img.height(), img.depth(), 3, 3);
img.blur_anisotropic(G);

Defensive patterns

Strategy: validation

Validate before calling

bool tensorOk = img.is_sameXYZ(G) && (G.spectrum()==3 || G.spectrum()==6);
if (!tensorOk) G.resize(img.width(), img.height(), img.depth(), 3, 3);

Type guard

bool validTensorField(const CImg<T>& img, const CImg<float>& G) {
  return img.is_sameXYZ(G) && (G.spectrum()==3 || G.spectrum()==6);
}

Try / catch

try {
  img.blur_anisotropic(G);
} catch (cimg_library::CImgArgumentException& e) {
  G.resize(img.width(), img.height(), img.depth(), 3, 3);
  img.blur_anisotropic(G);
}

Prevention

Build tensors (3 or 6 channels) from the same-resolution image
Never pass raw gradient stacks without checking spectrum count
Assert tensor shape in debug builds before heavy filtering

When it happens

Trigger: Passing a tensor field computed on a differently sized image; passing a gradient image with 1, 2 or more-than-6 spectra; passing an uninitialized or empty CImg as G.

Common situations: Computing structure tensors at a lower resolution for speed then forgetting to resize; constructing the tensor with Hessian components (4 channels in 2D) instead of the required 3; dimension mismatch after crop/resize.

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

AI-assisted analysis of Yalantis/uCrop@f788b534b4 (2026-09-08). Data as JSON: /api/errors/c324861f858f01bf. Report an issue: GitHub.