jolags313
commented
9 months ago Hello, could you please let me know how you managed to run the uHumans dataset?
I tried to use the kimera_semantics_uHumans2 launch file with one of the uHumans1 rosbags (retimestamped_humans_12_long.bag) and changed the inputs in the file to match those listed on the uHumans1 website, as well as rviz with kimera_semantics_gt.rviz, but it doesn't work.
All I see in rviz are the left image, depth, and segmentation feeds, and I am unable to see the reconstructed mesh representation/the map being built. The only thing that shows up are the coordinate frames on a black background.
Could you let me know the steps you took to run it?
Is the kimera-humans module implemented in the current state of the kimera-semantics package? In the paper, it is mentioned that dynamic masking is used to not perform person segmentation in kimera-semantics. However, when I executed the uhumans rosbag file with Kimera-Semantics, I was able to see human semantic meshes. Additionally, I noticed that smpl_female and smpl_male are mapped in the tesse_multiscene_office2_segmentation_mapping.csv file. Does this mean that kimera-semantics also performs human segmentation? If so, I would like to know which parts of the kimera-humans content mentioned in the paper are currently implemented in Kimera-Semantics.