在执行python finetune_autoencoder.py命令时报错，checkpoints文件夹为空

Question

在执行python finetune_autoencoder.py命令时报错，checkpoints文件夹为空

Closed this issue a year ago · 7 comments

在执行python finetune_autoencoder.py命令时报错，checkpoints文件夹为空。

finetune_autoencoder.py中的data_path = './data/mvtecad/'是不是要在根目录下再创建一个文件夹data/mvtecad/,那么在mvtecad文件夹下图片是如何存放的（目录是什么）

Answer 1 · 2024-01-02T13:50:33.000Z

Possibly you need to download the kl-f8.zip for the checkpoint. And you should create a path 'data/mvtecad/' where the directory is the same as the original MVTec-AD setting. Or you could change your dataset path in the finetune_autoencoder.py file. Or you could download our Pretrained Model.

|-- root
    |-- data
        |-- mvtecad
            |-- bottle
                |-- ground_truth
                    |-- broken_large
                        |-- 000_mask.png
                    |-- broken_small
                        |-- 000_mask.png
                    |-- contamination
                        |-- 000_mask.png
                |-- test
                    |-- broken_large
                        |-- 000.png
                    |-- broken_small
                        |-- 000.png
                    |-- contamination
                        |-- 000.png
                    |-- good
                        |-- 000.png
                |-- train
                    |-- good
                        |-- 000.png
        |-- train.json
        |-- test.json

Answer 2 · 2024-01-04T12:47:45.000Z

您好，根据您上述的解决方法做出改进后在执行python finetune_autoencoder.py命令时仍然报出相同的错误，之后我采用你们的预训练模型继续进行实验，在执行python build_model.py指令后可以正常生成diad.ckpt权重文件，但在执行python train.py时发生报错：

Answer 3 · 2024-01-04T13:36:45.000Z

It seems that your network cannot download the pretrained resnet50 model from timm, where in File "/root/DiAD-main/ldm/models/diffusion/ddpm.py", line 484, in on_validation_epoch_start
pretrained_model = timm.create_model("resnet50", pretrained=True, features_only=True).
One solution is that you could download the pretrained model first and put it in your path and revise the code:
pretrained_model = timm.create_model("resnet50", pretrained=False, checkpoint_path=your_path, features_only=True).

Answer 4 · 2024-01-05T01:10:43.000Z

从 timm 下载预训练 resnet50 模型，并修改路径如下代码：
pretrained_model = timm.create_model("resnet50", pretrained=False, checkpoint_path='/root/autodl-tmp/models/resnet50_a1_0-14fe96d1.pth', features_only=True)，运行python train.py时报错：

Answer 5 · 2024-01-05T02:25:38.000Z

try: pretrained_model = timm.create_model("resnet50", pretrained=False, checkpoint_path='/root/autodl-tmp/models/resnet50_a1_0-14fe96d1.pth', strict=False, features_only=True)

Answer 6 · 2024-01-08T10:07:35.000Z

修改为pretrained_model = timm.create_model("resnet50", pretrained=False, checkpoint_path='/root/autodl-tmp/models/resnet50_a1_0-14fe96d1.pth', strict=False, features_only=True)运行python train.py时报错：

Answer 7 · 2024-01-12T14:35:08.000Z

Or you could try to keep the original code and move the rresnet50_a1_0-14fe96d1.pth pretrained model to the path: "....../.cache/torch/hub/checkpoints"