The truthfulness of existing explanation methods in authentically elucidating theunderlying model’s decision-making process has been questioned. Existing meth-ods have deviated from faithfully representing the model, thus susceptible toadversarial attacks. To address this, we propose a novel eXplainable AI (XAI)method called SRD (Sharing Ratio Decomposition), which sincerely reflects themodel’s inference process, resulting in significantly enhanced robustness in ourexplanations. Different from the conventional emphasis on the neuronal level, weadopt a vector perspective to consider the intricate nonlinear interactions betweenfilters. We also introduce an interesting observation termed Activation-Pattern-Only Prediction (APOP), letting us emphasize the importance of inactive neuronsand redefine relevance encapsulating all relevant information including both activeand inactive neurons. Our method, SRD, allows for the recursive decomposition ofa Pointwise Feature Vector (PFV), providing a high-resolution Effective ReceptiveField (ERF) at any layer.