请问一下模型权重加载不成功是指定backbone，decoder错误吗？

[XIEFOX]

在encoder_decoder文件夹下执行test_gpu.py
准备好了处理好的potsdam数据集。
运行CUDA_VISIBLE_DEVICES=0 python "test_gpu.py" --backbone 'resnet50' --decoder 'upernet' --dataset 'potsdam' --ms 'False' --mode 'test' --resume "/media/sdb/xxxx/potsdam/resnet50_upernet_imp_sep_model.pth" --save_path "xxxxxx"
报错模型权重不匹配？
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for SemsegFinetuneFramework:
Missing key(s) in state_dict: "semseghead.1.weight", "semseghead.1.bias".
Unexpected key(s) in state_dict: "semseghead_1.1.weight", "semseghead_1.1.bias", "semseghead_2.1.weight", "semseghead_2.1.bias", "semseghead_3.1.weight", "semseghead_3.1.bias".

[DotWang]

@XIEFOX 匹配啊，这三个头只用于预训练，而不是微调，你测试不能直接拿预训练权重，而是拿微调后的权重

[XIEFOX]

请问运行best_internimage_t_upernet_finetune_model.pth应该用什么backbone，好像指定不了--backbone {resnet50,vitaev2,vitaev2_3,vit_b_rvsa}

[DotWang]

@XIEFOX 用internimage, model.py里边应该有选项

[XIEFOX]

（不好意思，我改错代码了）数据预处理我按https://github.com/open-mmlab/mmsegmentation/blob/master/docs/en/dataset_prepare.md#prepare-datasets2024-12-18 13:22:02,744 处理的，然后压缩包是5lable all和3ortho_irrg Testing time: 0:02:06
2024-12-18 13:22:02,745 Eval result: mIoU/mAcc/allAcc/mF1 0.0182/0.0314/0.0391/0.0434.
2024-12-18 13:22:02,745 Class_0 result: iou/f1/accuracy 0.0000/nan/0.0000, name: impervious_surface.
2024-12-18 13:22:02,745 Class_1 result: iou/f1/accuracy 0.0221/0.0432/0.0407, name: building.
2024-12-18 13:22:02,745 Class_2 result: iou/f1/accuracy 0.0023/0.0047/0.0034, name: low_vegetation.
2024-12-18 13:22:02,746 Class_3 result: iou/f1/accuracy 0.0645/0.1212/0.1106, name: tree.
2024-12-18 13:22:02,746 Class_4 result: iou/f1/accuracy 0.0022/0.0044/0.0024, name: car.这是不是有点低了?

[DotWang]

@XIEFOX 是的，你检查一下标签顺序吧，我这边的标签顺序和mmseg官方的不太一样

可以参考一下这个
#37