模型预训练权重在哪下载

[zgsxwsdxg]

你好！我在训练是发生找不到权重 VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch100/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth

[DotWang]

仓库主页

[zgsxwsdxg]

仓库主页

下载的文件名字是bit-rsp-vitaev2-s-levir-ckpt.pth和VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch100/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth 是一个吗？

[zgsxwsdxg]

仓库主页

if self.args.mode == 'imp': pretrained = '../pretrain_model/ViTAE_Window_NoShift_12_basic_stages4_14_best.pth.tar' elif self.args.mode == 'rsp_40': pretrained = '../VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch40/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth' elif self.args.mode == 'rsp_100':

pretrained = '../VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch100/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth'

        pretrained = 'D:\\codes\\ViTAE-Transformer-Remote-Sensing\\ChangeDetection\\weights\\levir\\3\\bit-rsp-vitaev2-s-levir-ckpt.pth'
    else:
        raise NotImplementedError

    if isinstance(pretrained, str):

        ckpt = torch.load(pretrained, map_location='cpu')

        if 'state_dict' in ckpt:
            _state_dict = ckpt['state_dict']

用下载那个权重bit-rsp-vitaev2-s-levir-ckpt.pth 训练时if 'state_dict' in ckpt: 报错了 提示： Traceback (most recent call last): File "train-tmp.py", line 103, in model = load_model(opt, dev) File "D:\codes\ViTAE-Transformer-Remote-Sensing\ChangeDetection\utils\helpers.py", line 188, in load_model with_pos='learned', enc_depth=1, dec_depth=8).to(device) File "D:\codes\ViTAE-Transformer-Remote-Sensing\ChangeDetection\models\networks.py", line 476, in init if_upsample_2x=if_upsample_2x, File "D:\codes\ViTAE-Transformer-Remote-Sensing\ChangeDetection\models\networks.py", line 386, in init norm_eval=False File "D:\codes\ViTAE-Transformer-Remote-Sensing\ChangeDetection\models\ViTAE_Window_NoShift\base_model.py", line 191, in init self.init_weights() File "D:\codes\ViTAE-Transformer-Remote-Sensing\ChangeDetection\models\ViTAE_Window_NoShift\base_model.py", line 210, in init_weights if 'state_dict' in ckpt: TypeError: argument of type 'BASE_Transformer' is not iterable

[zgsxwsdxg]

仓库主页

你base_model.py文件中有好多预训练权重路径，

    if self.args.mode == 'imp':
        pretrained = '../pretrain_model/ViTAE_Window_NoShift_12_basic_stages4_14_best.pth.tar'
    elif self.args.mode == 'rsp_40':
        pretrained = '../VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch40/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth'
    elif self.args.mode == 'rsp_100':
        # pretrained = '../VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch100/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth'
        pretrained = 'D:\\codes\\ViTAE-Transformer-Remote-Sensing\\ChangeDetection\\weights\\levir\\3\\bit-rsp-vitaev2-s-levir-ckpt.pth'
    else:
        raise NotImplementedError

我添加了下载的 pretrained = 'D:\codes\ViTAE-Transformer-Remote-Sensing\ChangeDetection\weights\levir\3\bit-rsp-vitaev2-s-levir-ckpt.pth'

[DotWang]

不是那个变化监测的主页，是总的主页

[DotWang]

https://github.com/ViTAE-Transformer/ViTAE-Transformer-Remote-Sensing

[DotWang]

变化检测分割之类的任务文件夹里边的模型是训练好的模型，不是预训练模型 预训练模型在本仓库主页，关于vitae，遥感这个仓库这里只提供了rsp-vitaev2-s-e100的 imp的在同项目的其他仓库中：https://github.com/ViTAE-Transformer/ViTAE-Transformer/tree/main/Image-Classification

[zgsxwsdxg]

项目

谢谢，rsp-vitaev2-s-e100 这个可以训起来了。 但是 if self.args.mode == 'imp': pretrained = '../pretrain_model/ViTAE_Window_NoShift_12_basic_stages4_14_best.pth.tar' 这个找不到时对应哪一个，仓库中：https://github.com/ViTAE-Transformer/ViTAE-Transformer/tree/main/Image-Classification有好多个，帮忙指点下，谢谢

[DotWang]

ViTAEv2-S那个

[zgsxwsdxg]

ViTAEv2-S那个

好了，可以了。
elif self.args.mode == 'rsp_40': pretrained = '../VitAE_window/output/ViTAE_Window_NoShift_12_basic_stages4_14_224/epoch40/ViTAE_Window_NoShift_12_basic_stages4_14/default/ckpt.pth' 这个是对应那个？

[DotWang]

这个不需要了，已经有epoch100的了，文章里边的实验已经表明40的不如100的，所以我就没放

[zgsxwsdxg]

这个不需要了，已经有epoch100的了，文章里边的实验已经表明40的不如100的，所以我就没放

ok，好的，谢谢。