jzhangbs / Vis-MVSNet

Visibility-aware Multi-view Stereo Network
MIT License
235 stars 27 forks source link

Performance - Tanks and Temple #9

Open eddienewton opened 3 years ago

eddienewton commented 3 years ago

Hi there,

Great work! I'm running your code on tanks and temple image sets.

I had a question about inference times. I'm currently getting about 9-10 seconds per image. Is this normal? I believe MVSNet and CasMVS are in the 1-2 second range right?

Just curious.

Thanks!

jzhangbs commented 3 years ago

it is possible. the time for pair-wise regularization is linear wrt number of views.