search

haofeixu / aanet

[CVPR'20] AANet: Adaptive Aggregation Network for Efficient Stereo Matching
Apache License 2.0
524 stars 102 forks source link

slightly confused about where the speed up came from #15

Closed hua-wu-que closed 4 years ago

hua-wu-que commented 4 years ago

Hi!

After reading your paper, I am not very clear how the speed up came from.

From my understanding, traditional cost volume approach build cost volume of size BXCXDXHXW, but in your approach you compute the dot product between two features, which results in BXDXHXW .

Is this where the speed up came from?

Thank you for your reply!

haofeixu commented 4 years ago

Cost aggregation on a 4D cost volume is significant faster than a 5D counterpart, you can check the Complexity Analysis part in Section 4.4 of our paper.