Support for raster BEV as fast rendering during logging #403
Conversation
| # Convert the BEV observation to a colored segmentation map | ||
| # If the tensor is one-hot encoded segmentation data | ||
| colored_bev = color_onehot_segmentation_map(bev, 'cpu') | ||
| agent_idx = 0 |
There was a problem hiding this comment.
This is always the sdc_index right? (I think so if I am following json_serialization.hpp well.
Maybe you would call this variable sdc_index just to it is clear we are logging that agent?
There was a problem hiding this comment.
yes, the 0th index is always the sdc (we follow json_serialization.hpp too)
Note: The implementation doesn't work yet: the raster image I get is black and white. Also, do you think it would be possible to display the raster image in the case when we control multiple agents?
There was a problem hiding this comment.
I think you need to permute the bev map so that the number of classes is the first dimension
bev.permute(2, 0, 1) or something like this.
For your second question, if im understanding correctly, I dont see an issue, you just might have to modify the colorization function to accept a first batch dimesion.
No description provided.