Questions about the different implementation of obs_encoder

[MsterDC]

Thanks to the author for providing the code, which is very extensible. But I have a question. Regarding the specific implementation of obs_encoder, I found that there are two implementation methods (standard resnet18 and SpatialSoftmaxPooling borrowed by robomimic), which was also mentioned in your previous answer. I want to know if there is any difference in performance improvement between these two encoders? In other words, which encoder is better?