I have similar problem in Issue 69 where the result of cpu version is different from gpu version. I check the answer in issue #69. And I use the original census transform that you provide in census_tansform.cu rather than symmetric census transform.
I also check the cost_aggregation.cu. I think there is nothing wrong in this part. In vertical part, update the vector by move block to left and down. In horizental part, update the vector by move the block to right. Thus, they are continus disparity vectors. But I can't get similar result in gpu version.
Can you give me some advice to solve the problem? Thank you.
I have similar problem in Issue 69 where the result of cpu version is different from gpu version. I check the answer in issue #69. And I use the original census transform that you provide in census_tansform.cu rather than symmetric census transform. I also check the cost_aggregation.cu. I think there is nothing wrong in this part. In vertical part, update the vector by move block to left and down. In horizental part, update the vector by move the block to right. Thus, they are continus disparity vectors. But I can't get similar result in gpu version.
Can you give me some advice to solve the problem? Thank you.