The paper addresses the issue of over-activation for head classes and under-activation for tail classes in Class Activation Mapping (CAM) generated through classifier weights in weakly supervised semantic segmentation (WSSS). This problem arises due to the long-tailed distribution in training data, where shared features between head and tail classes cause classifier weights to over-activate for head classes and under-activate for tail classes. To tackle this, the authors propose Shared Feature Calibration (SFC), a method that leverages class prototypes carrying positive shared features and introduces a Multi-Scaled Distribution-Weighted (MSDW) consistency loss to balance the activation of head and tail classes. The MSDW loss counteracts over-activation and under-activation by calibrating shared features in classifier weights. Experimental results on Pascal VOC 2012 and COCO 2014 datasets demonstrate that SFC significantly improves CAM boundaries and achieves state-of-the-art performance in WSSS. The project is available at https://github.com/Barrett-python/SFC.The paper addresses the issue of over-activation for head classes and under-activation for tail classes in Class Activation Mapping (CAM) generated through classifier weights in weakly supervised semantic segmentation (WSSS). This problem arises due to the long-tailed distribution in training data, where shared features between head and tail classes cause classifier weights to over-activate for head classes and under-activate for tail classes. To tackle this, the authors propose Shared Feature Calibration (SFC), a method that leverages class prototypes carrying positive shared features and introduces a Multi-Scaled Distribution-Weighted (MSDW) consistency loss to balance the activation of head and tail classes. The MSDW loss counteracts over-activation and under-activation by calibrating shared features in classifier weights. Experimental results on Pascal VOC 2012 and COCO 2014 datasets demonstrate that SFC significantly improves CAM boundaries and achieves state-of-the-art performance in WSSS. The project is available at https://github.com/Barrett-python/SFC.