Multiclass Segmentation Performance Metrics

[VAM7686]

trafficstars

This is what the multiclass segmentation output currently looks like:

• https://www.dropbox.com/s/yuv4jou4ch4uqla/multiclassSegmentation.webm?dl=0

The output looks a bit choppy and I'm not sure if this is due to the GPU the model ran on, the node that runs model, or even the frame rate of the ros bag file that was used.

It would be nice to collect metrics on the new multiclass segmentation model and how it compares to our old binary segmentation model.

Thoughts:

• Check frame rate of the segmentation topic
• Check model inference rate.
• If the model inference rate of binary and multiclass are not similar, perhaps check the image frame rate or other part of code that slows the code

AC: Performance metrics of multiclass segmentation versus binary segmentation which can be included in the documentation. Possible improvements to inference rate of multiclass segmentation.