search

personalrobotics / OpenChisel

An open-source version of the Chisel chunked TSDF library.
459 stars 128 forks source link

toooooo slow test on the Freiburg RGBD dataset #13

Open ayumizll opened 8 years ago

ayumizll commented 8 years ago

hi: i test chisel on my desktop equipped with Intel® Xeon(R) CPU E3-1231 v3 @ 3.40GHz × 8, and i run the launch_freiburg_dataset.launch with fault parameters . every update frame i can get nice reconstruction result, but too slow( Almost 3 seconds). i have no idea if there are some important parameters need to be set, or some significant things to be all attention. thanks!

mklingen commented 7 years ago

If you run the bag file without running chisel, how fast does it play back? Also, are you compiling in Release mode?

ayumizll commented 7 years ago

hi mklingen，Thank you for your answer. I'm sure on the release model, and I think this is independent of
ROS bag time. I test the chisel, find the most time-consuming is the function int Chisel.h named IntegrateDepthScanColor,

mklingen commented 7 years ago

I made some changes recently which should make IntegrateDepthScan a bit faster. I will push these sometime this week, and I'll let you know. At the time when I was running this on the Freiburg dataset I was getting ~10Hz out of it in Release mode. (-DCMAKE_BUILD_TYPE=Release).

The parameters which will affect the performance the most are voxel resolution (voxel_resolution_m) and the frustum cutoff (far_plane_dist) use_voxel_carving and to a lesser extent use_color will also make it slower.

ayumizll commented 7 years ago

hi mklingen, i am looking forward to your updated version. I will try to change the parameters and test the time again.

HubertYang commented 7 years ago

In my test, Chunk's size is also important. When resolution is 5cm, it is the fastest that chunk's size is 4x4x4. Each resolution, you should find the best chunk's size.