gasharper / pyramidflow Goto Github PK

Hello，your paper is excellent. What is this project's [license]？MIT?

What is the reason why the Image-level AUROC metric and the Pixel-level pAUROC metric are particularly low, only 0.575 and 0.872 respectively, when running the screw category with default settings?

nvrtc: error: failed to open nvrtc-builtins64_118.dll.
Make sure that nvrtc-builtins64_118.dll is installed correctly.

Thank you for opening the source code. Can you provide the code of model inference? My email is [email protected]

Hello, could you please publish the code for visualizing the localization results?

您好！我对您团队的研究非常感兴趣，并觉得十分有意义。但是我目前有一个疑问，既然模型使用正常样本图像进行训练。为什么数据集中还用到了类似labelme标注完成的ground truth。 ground truth在文中起到了什么作用，是最后的测试结果的精度评估吗？是否涉及了训练过程？ 非常感谢并期待您的回复！

When the train code runs to the file "model.py", I encountered the above error. In this file, it has “from autoFlow import SequentialNF, InvertibleModule, SequentialNet”.

Is this method only using defect free samples for template comparison during the training phase?

Hi,

I believe your implementation of the AUPRO score is missing an important part (or at least i couldnt find it).

https://github.com/gasharper/PyramidFlow/blob/6977d5a8294276bf7a9952477235f219484c2218/util.py#L156C20-L156C20

The usual and recommended way to compute it is by cutting off the PRO curve at 30% on the x-axis (FPR), then taking the area under the curve to the left of that point and normalizing the score (divide by 30%).

An extract:

from

[1] P. Bergmann, K. Batzner, M. Fauser, D. Sattlegger, and C. Steger, “The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection,” IJCV, vol. 129, no. 4, pp. 1038–1059, Apr. 2021, doi: 10/gjp8bb.

This has an important implication because not cutting the curve inflates the results significantly.

Here is a reference a implementation:

https://github.com/jpcbertoldo/anomalib/blob/b4f166d9b5c7efeb6e013c5c3cf25fecf6156b04/src/anomalib/utils/metrics/pro.py#L17

您好，我在安装您指定的库时，报错，请问您的环境配置是什么啊？比如windows还是ubuntu，cuda和torch是哪个版本的啊？期待您的回复，谢谢！

您好，在train过程中，加载数据集遇到 “AttributeError: Can't pickle local object 'fix_randseed..seed_worker'” 的问题，请问您有遇到过吗？

您好，我在复现本项目代码时，在仅仅修改了datapath的情况下，运行train.py，就出现了AttributeError: Can't pickle local object 'fix_randseed..seed_worker'问题。我将问题锁定在 train.py 的 line44 line45，但我对解决这个问题无能为力。请您们确认您们的代码是准确无误且可移植的。

以下是我的代碼：

checkpoint = torch.load(save_name, map_location=torch.device('cpu')) # 加载模型参数

resnetX = checkpoint['resnetX']
num_layer = checkpoint['num_layer']
vn_dims = checkpoint['vn_dims']
ksize = checkpoint['ksize']
channel = checkpoint['channel']
num_stack = checkpoint['num_stack']
batch_size = checkpoint['batch_size']
state_dict_pixel = checkpoint['state_dict_pixel']

初始化 PyramidFlow 模型

flow = PyramidFlow(resnetX=resnetX, num_level=num_layer, vn_dims=vn_dims,
ksize=ksize, channel=channel, num_stack=num_stack)

flow.load_state_dict(state_dict_pixel)#此處報錯

報錯內容如下：

RuntimeError: Error(s) in loading state_dict for PyramidFlow:
size mismatch for nf.moduleslst.0.affineParams.norm.running_mean: copying a param with shape torch.Size([1, 1, 128, 128]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).
size mismatch for nf.moduleslst.1.affineParams.norm.running_mean: copying a param with shape torch.Size([1, 1, 64, 64]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).
size mismatch for nf.moduleslst.2.affineParams.norm.running_mean: copying a param with shape torch.Size([1, 1, 256, 256]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).
size mismatch for nf.moduleslst.3.affineParams.norm.running_mean: copying a param with shape torch.Size([1, 1, 64, 64]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).
size mismatch for nf.moduleslst.4.affineParams.norm.running_mean: copying a param with shape torch.Size([1, 1, 128, 128]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).
size mismatch for nf.moduleslst.5.affineParams.norm.running_mean: copying a param with shape torch.Size([1, 1, 64, 64]) from checkpoint, the shape in current model is torch.Size([1, 1, 1, 1]).

參數皆是設置為預訓練好的模型參數，但仍然報錯。
是否參數內容設置錯誤，再勞煩指教，感謝！

您好，在研读您的论文时发现在多个数据集都有很好的效果，但是代码中似乎只有针对mvtec的训练等代码，可以提供一下有关btad数据集的训练等代码吗？谢谢

https://github.com/gasharper/PyramidFlow/blob/c463b1cb0c2b084cdbae290234e3f8657298e981/model.py#L94C14-L94C14
running_mean 不能加载