When running Pylot with the perfect detector operator on obstacles (such as in the perfect_detection.conf), I noticed that the detector often misses bicycles (even when the bicycle is very close to the ego vehicle) and occasionally some other vehicles.
I think that the self.__segmentation_threshold (line 59) field in the obstacle.py file is likely causing the missing bicycles in the detector when it is being used to populate the 2D bounding boxes for the perfect detector’s obstacles. Currently, the threshold value is 0.20 and causes a lot of the missed/flickering detections, but we noticed that changing it to a lower value like 0.01 causes the detector to consistently detect all of the bicycles.
Here is an example of the detector not detecting bicycles when the threshold=0.20:
Here is an example of the detections when the threshold=0.01:
The text and color of the bounding boxes might be different since I was modifying that part a bit before discovering this issue.
Hello,
When running Pylot with the perfect detector operator on obstacles (such as in the perfect_detection.conf), I noticed that the detector often misses bicycles (even when the bicycle is very close to the ego vehicle) and occasionally some other vehicles.
I think that the self.__segmentation_threshold (line 59) field in the obstacle.py file is likely causing the missing bicycles in the detector when it is being used to populate the 2D bounding boxes for the perfect detector’s obstacles. Currently, the threshold value is 0.20 and causes a lot of the missed/flickering detections, but we noticed that changing it to a lower value like 0.01 causes the detector to consistently detect all of the bicycles.
Here is an example of the detector not detecting bicycles when the threshold=0.20:
Here is an example of the detections when the threshold=0.01:
The text and color of the bounding boxes might be different since I was modifying that part a bit before discovering this issue.
Just wanted to give a heads up -- Thanks!