ConvX: Efficient XAI for real-time computer vision with single forward-pass explanations

In computer vision, many of the popular techniques for explainable AI (XAI) in multiclass classification are based on post-hoc analysis using computationally intensive backward passes through a network, e.g., Grad-CAM. In this paper, we introduce a fully convolutional method for explainable real-time image classification that operates with a high-speed single forward pass for all classes: ConvX. To accomplish this, we reformulate visual attribution as a multiclass multilabel hypersphere classification problem, embedding the attribution mechanism directly into the forward inference pathway of the network. In addition, our framework introduces an optional hybrid supervision mechanism where the system can accommodate a small subset of pixel-level masks into the loss function to enforce spatial regularization. Benchmarking results on the publicly available PASCAL VOC 2012 dataset show that our proposed method achieves a mean Intersection over Union (mIoU) of 0.34 using only image-level labels in a flat, deterministic runtime of under 10 ms. Our proposed method provides comparable localization performance to post-hoc alternatives such as Grad-CAM (0.35 mIoU), which takes double the time to process one class and slows down linearly as the number of classes increases. Furthermore, using just five ground-truth masks per class, our hybrid variant achieves an mIoU of 0.48, substantially outperforming Grad-CAM and variants while maintaining the same low latency as ConvX. These results demonstrate that our framework is highly suited for deployment in edge environments such as robotics and industrial automation.