tuantng / temporalmaxer Goto Github PK

Hello, I would like to know the configuration parameters of multiTHUMOS that you used in ActionFormer when you were doing the comparison experiment. Can you send a copy parameter?

自己的数据怎么提pkl文件

Very simple but effective model! Howerver, I notice you use SimOTA in your code, but Actionformer does not use it. So I want to know if SimOTA can bring some performance improvement.

When first training THUMOS, an AttributeError was raised:

AttributeError: module 'numpy' has no attribute 'float'

Now we often use 'np.float32' with the fact that 'np.float' has been deprecated in NumPy. But it still exists in 'libs/utils/metrics.py'.

I performed scripts/multithumos/train.sh and test.sh to reproduce the experiment, but it can't achieve the official performance

above is my reproduce result, how should I adjust the ymal setting to achieve the official result

Thanks