Visual classifiers trained with standard augmentation pipelines tend to exploit spurious correlations:background statistics, texture patterns, or scene-level cues that happen to co-occur with class labelsbut do not reflect object structure. Conventional augmentation strategies, including random crops,colour jitter, and policy-based methods such as RandAugment and AutoAugment, are constructedwithout reference to what the model has actually learned, and so offer no mechanism for targeting thespecific shortcuts a given classifier has acquired.We present Intelligent Coarse Dropout (ICD) and its complement Anti-ICD (AICD), two dataaugmentation methods that condition spatial masking on the model’s own class activation maps.Standard coarse dropout (cutout) hides rectangular regions chosen at random, erasing background asreadily as object. ICD and AICD make this saliency-aware: they split the image into a grid of tiles,score each tile by its mean saliency, and selectively mask tiles by a percentile threshold. ICD masksthe highest-saliency tiles, the regions the model already relies on, so that the classifier must recruitadditional cues. AICD masks the lowest-saliency tiles, perturbing the background while leaving thediscriminative region intact. Masked tiles are not filled with a hard black box but with one of severalsoft strategies (blurred original, local mean, noise, or a constant), which preserve different aspects oflocal context.We describe the formal construction of both masks, the family of fill strategies, the role of thetile size and threshold hyperparameters, and a reference implementation in the open-source BNNRlibrary. This paper introduces the methods; a quantitative evaluation against standard augmentationbaselines is left to a companion study.