Test: poor depth output with CPU

[OceanYing]

Hi! I want to run the test code with a CPU and now there is no error appears during the test. But I found the depth outputs (flow1,2,3) are too poor to get good fusion result. Below is a example (scene1) The init, flow1,2,3 looks similar.

The camera parameter comes from the training data of MVSNet, and I use DTU rectified scene1 as input. I just modified the file path of the original code. I tried with my GPU(4GB) but there is an OOM error.

Could you give me some advice?