mx.image.imdecode to numpy causes the speed to slow down

[hch-baobei]

Now I have encountered such a problem that I use opencv and pil to read pictures slowly, so I chose mxnet, but after I changed mxnet.image.imdecode into numpy format (here is for subsequent image processing), The speed of the three is about the same. So I would like to ask if there is any win-win method, which can not only speed up, but also facilitate the subsequent image processing. Great community friends, please give me some ideas.
this is slow

this is fast