Does I have to download the imagenet dataset for training the model ? when i run train_reduced_student.py ...
thanks !

Thanks for your contribution

Kind regards

Can I perform model distillation training with any img size not just 256*256? It cause errors that tensor size not match.

Hello, Thank you for your work. I downloaded the pre-trained weight from the provided link. https://github.com/rximg/EfficientAD/releases/tag/0.0.1

The link contains the weights of only the teacher network. But from where I can get pre-trained weight for the student and autoencoder network? Thank you

batch response: This repository is over its data quota. Account responsible for LFS bandwidth should purchase more data packs to restore access.
error: failed to fetch some objects from 'https://github.com/rximg/EfficientAD.git/info/lfs'

Please ask! What is the final output obtained by converting the trained model into onnx format and importing it into onxxruntime for inference? Is it the input sample detection result (and whether it is an anomalous sample) and the corresponding confidence level? Or is it the anomaly_map with the same channel 1 as the input height and width?

Hi! Because of my old device, could you provide pretrain student's and ae's ckpt?
Thanks a lot!

How should I start training?

推理出的热力图为什么4个角会显示异常

When I tried git lfs pull, i saw below message.

batch response: This repository is over its data quoa.
Account responsible for LFS bandwidth should purshase more data packs to restore access.
Failed to fetch some objects from 'https://github.com/rximg/EfficientAD.git/info/lfs'

Could i get data using other ways?

Thank you!

I have run the pretraining and got the same results as in the paper. You can get the pretrained weights here: https://github.com/nelson1425/EfficientAD

distillation_training.py", line 100, in global_channel_normalize
ldist = item['image'].cuda()
IndexError: too many indices for tensor of dimension 4

请检查提交的代码是否OK？，还是我数据集存在问题？ 请帮忙指点
ImageNet
-train
-n01440764
-....
-val
-test

Hi, here are some qualitative and quantitative results on MVTec bottle defect of the EfficientAD(M) method.

image_AUROC: 1.0
image_F1Score: 1.0
pixel_AUROC: 0.9876494407653809
pixel_F1Score: 0.7927650213241577

Patchcore on Wide ResNet-50 achieves image_AUROC: 1.0 and pixel_AUROC: 0.984

should i train teacher model on my own datasets with anomaly segment mask and train student model and ae model only on normal sample on my dataset