Some questions about “PConv + PWConv“
LKAMING97 opened this issue · 5 comments
I would like to know if PConv + PWConv, described in the thesis, can also be realized like T-shape Conv, so that more attention is paid to the central position, or it can only be used to let all information flow through all channels？
@LKAMING97 Hi, T-shaped Conv requires non-trial implementation and thus we adopt out-of-the-box PConv and PWConv. The combination of PConv and PWConv also has lower FLOPs and fewer parameters compared to the T-shaped Conv.
Hello,I have a question here. If I want to apply the model to the classification of one-dimensional signals, whether the convolution is changed to 1d, and the effect of partial convolution can also be achieved.
@LKAMING97 Hi, the PConv can be changed into 1D, whose effectiveness depends on the input redundancy of your task.
Hello, I just looked at the code, and I found out why "fuse_conv_bn" is not needed in the "evaluation command", but it is used when measuring the latency. Will there be any deviation?
@LKAMING97 Hi, the "evaluation command" you mentioned refers to the evaluation of performance, e.g., accuracy, regardless of the latency. Therefore, "fuse_conv_bn" is not compulsory. You may also turn it on and the accuracy is almost the same.