The field of computer vision is witnessing significant advancements in weakly supervised learning, with a focus on developing methods that can learn from limited annotated data. Researchers are exploring innovative approaches to address the challenges of fine-grained visual categorization, object segmentation, and defect detection. A common theme among these approaches is the use of pseudo-labels, soft labels, and contrastive learning to improve model performance. Notably, the integration of region-aware attention, Bayesian uncertainty, and diffusion features is showing promising results. Noteworthy papers include RAUM-Net, which proposes a semi-supervised method for fine-grained visual categorization, and Weakly Supervised Object Segmentation by Background Conditional Divergence, which introduces a method for training a masking network using weak supervision. Region-Aware CAM and Contrastive Learning with Diffusion Features are also making significant contributions to the field of weakly supervised semantic segmentation.