mingo_敏

为了解决模型预测的铁轨掩码边缘粗糙的问题，本文将边缘算子提取的边缘信息融入原始 Yolact 的损失函数中，以强调模型对铁轨边缘的关注。此外，本文应用盒式滤波器对线性插值导致的真实标签掩码边缘锯齿进行平滑处理。由于边缘信息的融入和平滑处理过程仅在训练阶段进行，因此模型的推理速度并未受到影响。本文在自定义铁轨数据集上的实验结果显示，预测准确率有所提高。

深度学习论文: Cultivated Land Extraction from High-Resolution Remote Sensing ImageThe Winning Solution to the iFLYTEK Challenge 2021 Cultivated Land Extraction from High-Resolution Remote Sensing ImagePDF: https://arxiv.org/pdf/2202.10974.pdfPyTorch代码: https:

CenterMask : Real-Time Anchor-Free Instance SegmentationPDF:https://arxiv.org/pdf/1911.06667.pdfPyTorch: https://github.com/shanglianlm0525/PyTorch-Networks1 概述在FOCS检测的基础上通过添加一个SAG-Mask（spatial attention-guided mask）分支来实现像Mask RCNN中类似的分割效果。2 CenterMas

Key Points Estimation and Point Instance Segmentation Approach for Lane DetectionPDF: https://arxiv.org/pdf/2002.06604.pdfPyTorch代码: https://github.com/shanglianlm0525/PyTorch-NetworksPyTorch代码:...

PolarMask: Single Shot Instance Segmentation with Polar RepresentationPDF: https://arxiv.org/pdf/1909.13226.pdfPyTorch: https://github.com/xieenze/PolarMaskPyTorch: https://github.com/shanglianlm05...