EfficientAd is slower than other models in anomalib

[haimat]

I had the impression that the EfficientAd model would be among the fastest in anomalib in terms of prediction times. To verify that I have trained three models, Padim, Fastflow, and EfficientAd, all with the same training data and an image dimension of 512x512 pixels. Then I have written a small script that loads these models, warms up the GPU, and then runs prediction on 100 images. I measure only the model forward time, no image loading or any pre- or post-processing.

With the models exported to ONNX I get these results (avg. model forwards times on 100 images):

• EfficientAd: 0.0116 sec.
• Fastflow: 0.0053 sec.
• Padim: 0.0036 sec.

So in other words: The EfficientAd model is the slowest from these three, and Padim the fastest - I thought it would be the other way round. Am I missing something, or is this a bug in anomalib?

[haimat]

Hello, no news on this?

[Iheb-prepa]

To my knowledge, there's no official code for EfficientAd, however you can check this unofficial implementation which claims to match the paper's results.

[abc-125]

Hello, I have a couple of suggestions. You can try 256x256 pixels size to match papers. Also, you can try to run it without ONNX, optimization might be working differently for different models.

[Issa-Bqain]

@haimat Can you please share the source code you are using for this comparison and I can see if these results are reproducible on my end? thanks

[haimat]

@haimat Can you please share the source code you are using for this comparison and I can see if these results are reproducible on my end? thanks

Unfortunately not, I don't have the code any more, it was an ad-hoc test.
But quite easy to reproduce, as described in my original posting.