SasWOT is a training-free semantic segmentation architecture search (SAS) framework that uses an auto-discovered proxy to efficiently search for optimal segmenters. Unlike traditional SAS methods that require training, SasWOT leverages evolutionary algorithms to search for proxies without training, significantly improving search efficiency. The framework designs a proxy search space based on statistical inputs from segmenters and uses mathematical operations to represent candidate proxies. It then employs an evolutionary algorithm to evolve these proxies, using ranking correlation with segmentation benchmark results as the optimization objective. This approach allows SasWOT to find effective proxies for segmenter search without any training cost. The search process is accelerated by using judgment and elitism-preserve strategies during proxy search. SasWOT achieves a superior trade-off between accuracy and speed compared to several state-of-the-art techniques. On the Cityscapes dataset, SasWOT achieves a mIoU of 71.3% with a speed of 162 FPS. Extensive experiments on Cityscapes and CamVid datasets demonstrate that SasWOT significantly improves ranking consistency compared to other training-free methods. It also achieves a search acceleration of at least 30 times compared to traditional gradient-based SAS methods, allowing segmenter search to be completed on a single GPU within 2 hours. The main contributions of SasWOT include proposing a novel training-free SAS framework, presenting a comprehensive proxy search space with correlation-based fitting objectives, and achieving significant search acceleration through discovered proxies. SasWOT also conducts extensive experiments on standard benchmarks, showing superior performance and efficiency. The framework is designed to automatically discover proxies for SAS tasks, enabling efficient and effective segmenter search without training. The results demonstrate that SasWOT is a promising approach for real-time semantic segmentation, offering significant improvements in search efficiency and performance.SasWOT is a training-free semantic segmentation architecture search (SAS) framework that uses an auto-discovered proxy to efficiently search for optimal segmenters. Unlike traditional SAS methods that require training, SasWOT leverages evolutionary algorithms to search for proxies without training, significantly improving search efficiency. The framework designs a proxy search space based on statistical inputs from segmenters and uses mathematical operations to represent candidate proxies. It then employs an evolutionary algorithm to evolve these proxies, using ranking correlation with segmentation benchmark results as the optimization objective. This approach allows SasWOT to find effective proxies for segmenter search without any training cost. The search process is accelerated by using judgment and elitism-preserve strategies during proxy search. SasWOT achieves a superior trade-off between accuracy and speed compared to several state-of-the-art techniques. On the Cityscapes dataset, SasWOT achieves a mIoU of 71.3% with a speed of 162 FPS. Extensive experiments on Cityscapes and CamVid datasets demonstrate that SasWOT significantly improves ranking consistency compared to other training-free methods. It also achieves a search acceleration of at least 30 times compared to traditional gradient-based SAS methods, allowing segmenter search to be completed on a single GPU within 2 hours. The main contributions of SasWOT include proposing a novel training-free SAS framework, presenting a comprehensive proxy search space with correlation-based fitting objectives, and achieving significant search acceleration through discovered proxies. SasWOT also conducts extensive experiments on standard benchmarks, showing superior performance and efficiency. The framework is designed to automatically discover proxies for SAS tasks, enabling efficient and effective segmenter search without training. The results demonstrate that SasWOT is a promising approach for real-time semantic segmentation, offering significant improvements in search efficiency and performance.