search

HuiZhang0812 / DiffusionAD

148 stars 16 forks source link

Experiment on MVTec LOCO #17

Closed Ace-blue closed 10 months ago

Ace-blue commented 10 months ago

Dear @HuiZhang0812 , your work is amazing. Now I want to try DiffusionAD on dataset MVTec LOCO. But I donot know if your method can work on this dataset. Have your team ever tried this dataset? Any suggestions for me if I want try?

Thanks for your reply ^-^

HuiZhang0812 commented 10 months ago

I have not yet tried the MVTec LOCO dataset. However, I believe that for any dataset, as long as you design a suitable anomaly synthetic strategy (such as object foreground, etc.), DiffusionAD can be deployed.

Airbyj92 commented 10 months ago

Is the png image in the thresh subfolder a binary graph of the positive sample in training?

HuiZhang0812 commented 10 months ago

Is the png image in the thresh subfolder a binary graph of the positive sample in training?

yes!

Airbyj92 commented 10 months ago

thresh 子文件夹中的 png 图像是训练中阳性样本的二进制图吗？

是的！

Hello, I use 4090 inference speed of 550 ms/piece please ask the speed can also be increased

HuiZhang0812 commented 10 months ago

thresh 子文件夹中的 png 图像是训练中阳性样本的二进制图吗？

是的！

Hello, I use 4090 inference speed of 550 ms/piece please ask the speed can also be increased

Does the inference time you calculated include the computation time for metrics like Pro?

Airbyj92 commented 10 months ago

thresh 子文件夹中的 png 图像是训练中阳性样本的二进制图吗？

是的！

您好，我使用4090推理速度为550毫秒/件，请询问速度也可以提高

您计算的推理时间是否包括 Pro 等指标的计算时间？

Using your original prediction code

HuiZhang0812 commented 10 months ago

thresh 子文件夹中的 png 图像是训练中阳性样本的二进制图吗？

是的！

您好，我使用4090推理速度为550毫秒/件，请询问速度也可以提高

您计算的推理时间是否包括 Pro 等指标的计算时间？

Using your original prediction code

The eval.py script does not involve the calculation of frames per second (FPS) yet. Inference speed varies across different GPUs, and I believe that computation time for performance metrics should not be included when calculating inference time. How did you obtain the 550 ms/piece inference speed?​

Airbyj92 commented 10 months ago

thresh 子文件夹中的 png 图像是训练中阳性样本的二进制图吗？

是的！

您好，我使用4090推理速度为550毫秒/件，请询问速度也可以提高

您计算的推理时间是否包括 Pro 等指标的计算时间？

Using your original prediction code

The eval.py script does not involve the calculation of frames per second (FPS) yet. Inference speed varies across different GPUs, and I believe that computation time for performance metrics should not be included when calculating inference time. How did you obtain the 550 ms/piece inference speed?​

There will be a rate of inference per second, but there should be some metric calculation mixed in