MNN Version 73ad3413 is significantly slower in execution on the opencl backend than its predecessor, cfad1c3

u3r8eeie  于 2021-12-26  发布在  Java
关注(0)|答案(0)|浏览(212)

I used PowerVR GX6650 to test deeplabv3 model in the demo. Before, it took 72ms to run the low-precision mode, but now it takes 86ms, and my own model took 375ms before, but now it takes 435ms.

暂无答案!

目前还没有任何答案,快来回答吧!

相关问题