search

MIT-SPARK / Kimera-Semantics

Real-Time 3D Semantic Reconstruction from 2D data
BSD 2-Clause "Simplified" License
631 stars 134 forks source link

Caught an unknown color #45

Closed p-akanksha closed 3 years ago

p-akanksha commented 3 years ago

Hi,

I am trying to run Kimera Semantic on the data collected from the Tesse simulator (the one used in the goseek-challenge). I keep getting the following error: 

E1027 12:28:32.195211 24858 color.cpp:73] Caught an unknown color: 
RGB: 143 171 0 255

Also, the reconstructed point cloud does not have associated semantic labels. Here is an image of the point cloud. Kiemra_color_problem

I am unable to figure out what might be causing this error. Any help is highly appreciated.

ToniRV commented 3 years ago

Hi @p-akanksha,

There is a .csv file inside the cfg folder (configuration) that specifies the mapping between colors and object ids.

Depending on the colors in the semantic image, you might need to specify one csv vs another. I think the one for go-seek should be already the default one: https://github.com/MIT-SPARK/Kimera-Semantics/blob/7198e5a3998321671ace2cbafe0f8a640f07cac0/kimera_semantics_ros/launch/kimera_semantics.launch#L126-L127

But maybe you can try replacing office1 in the line above by office2?

Ping @ZacRavichandran

ZacRavichandran commented 3 years ago

Hi @p-akanksha

Like @ToniRV mentioned, the error indicates that there was an object not specified in the configuration .csv. There's a configuration for the goseek-challenge scenes, goseek_segmentation_mapping_dense.csv, that should contain all the necessary information. Perhaps you could try replacing tesse_multiscene_office1_segmentation_mapping.csv with that?

p-akanksha commented 3 years ago

Thanks! Replacing tesse_multiscene_office1_segmentation_mapping.csv with goseek_segmentation_mapping_dense.csv worked perfectly.

This has also increased the speed of 3d reconstruction. Earlier it felt like it was processing frames in batches, now it reconstructs immediately. Do you think the delay was because it was trying to look for the objects in the configuration file and unable to find them?

ToniRV commented 3 years ago

Hi @p-akanksha,

Glad it worked!

The lookup of the semantic id for a given color should have a constant time complexity as we are using std::unordered_map. Nevertheless, if the getSemanticLabelFromColor function doesn't find a color, it will start printing errors with the missing colors, this is not particularly fast.

p-akanksha commented 3 years ago

That makes sense. Thanks for the prompt response.