Closed rkuo2000 closed 3 years ago
The cause is I directly use pretrain model to predict without training the pretrain model. model_name = "pspnet_mobilenet_v2" has much better result than "fcn8_vgg16" rescheduler did not help trainer (it was commented anyway)
https://kaggle.com/rkuo2000/semantic-segmentation-pytorch for your reference !
predict.py ---> 81 model_width = model.img_width 82 model_height = model.img_height
ModuleAttributeError: 'FCN8' object has no attribute 'img_width'