This warehouse mainly uses two model technologies, DeepLabV3 and DeepLabV3Plus, and mixes MobileNetV2, ResNet101, and HRNetV2 to implement segmentation tasks. Slightly modified based on the source code. You are free to use VOC, CityScapes datasets, and your own datasets to train any segmentation models. Download links for the VOC and CityScapes datasets are below.
Download VOC and CityScapes： 链接：https://pan.baidu.com/s/1fy0R6NJo0YanfD4n6kgCiw?pwd=0grl 提取码：0grl --来自百度网盘超级会员V3的分享
Note: Before training the model, you need to enter the train_gpu.py file to modify the path and batchsize of your own data set. If you enable automatic mixed precision and then load the weight file under the checkpoints file, you may(sometimes) encounter a ComplexFloat error during the forward process. (If you know how to solve this problem, you are welcome to suggest a solution, thank you very much!)
- cd DeepLabV3Plus
- python train_gpu.py