Wild-scene-image-segmentation(2D only)

The project collected more than 5,000 wild environment images and corresponding labels. First, the DeepLabV3-ResNet50 method in the paper K. Vidanapathirana et al. (2023). WildScenes: A Benchmark for 2D and 3D Semantic Segmentation in Large-scale Natural Environments. was reproduced. Then the team innovated two methods, DeepLabV3-MobileNetV3 and SegNet-ResNet50, to train and predict the data set, and drew an accuracy graph to find out the advantages and disadvantages of each method. Finally, the innovative method was optimized to keep its iou above 40%, and a paper was written. Webpage: WildScenes: A Benchmark for 2D and 3D Semantic Segmentation in Natural Environments. CSIRO 2023. https://csiro-robotics.github.io/WildScenes/ Paper: K. Vidanapathirana et al. (2023). WildScenes: A Benchmark for 2D and 3D Semantic Segmentation in Large-scale Natural Environments. arXiv:2312.15364. https://arxiv.org/abs/2312.15364 Dataset: CSIRO Data Access Portal: WildScenes Dataset. Version 2 (12 June 2024). https://doi.org/10.25919/5hzc-5p73