Open fabfabBesnard opened 4 years ago
This one will not be easy to achieve with a clean solution. The solution that might be the cleanest is locating how transforms are deduced and working from there because looking at the clouds from two datasets with v1 and v2 robots, the orientation is the same. Which means somewhere a wrong transform is applied. Finding where might allow us to detect from which version the datasets come from. A dirty-er solution would be flipping the cloud in the vertex shader and would require information on the dataset level. Proper time should be dedicated to investigate this issue.
Intuition : Maybe check the way cameras are loaded. It may impact the global transform of the scene. Try loading in the point cloud first if it exists and see if the issue still happen
tested with :
roboticsmicrofarms/romiscan demo 5161bfd41de9