Modality-Resilient Multimodal Industrial Anomaly Detection via Cross-Modal Knowledge Transfer and Dynamic Edge-Preserving Voxelization

. 2025 Oct 23;25(21):6529. doi: 10.3390/s25216529

Algorithm 1: Algorithm of the proposed method for two modalities.

Inputs : D^{1 c}, D^{2 c}

d^{c}

D^{1 r}

d^{1 u}

D^{2 p}, d^{2 p}

α, β

for each training iteration do
Train teacher model

L_{{Te}_{1}}

with {[

D^{1 c}

D^{1 r}

], [

d^{c}

d^{1 u}

]}
end for
for each training iteration do
Train teacher model

L_{{Te}_{2}}

with {[X2c, X2u], [

d^{c}

d^{2 p}

]}
end for
Label the samples to train student model with the teachers with Equation (14).
for each training iteration do
Train student model with Equation (15)
end for