It's been noticed that the detection of the busses is not close to 100%. While it is picking up a lot of busses, a large proportion are not being detected at all (the yellow box around them in debug screen). This can lead to false negatives, which is hard to detect without manual intervention.
Some of the cases are probably due to:
The ledge
The tree
The traffic light pole
Trucks parking at bottom of frame
Busses in far lane obscured by vehicles in close lane
These will probably require retraining based on old footage, and/or on new footage.
Occasionally some false positives are noted due to people walking across road (and other causes not saved on video due to green light). These are in general more noticed due to the positive nature meaning that a video is saved and can be used for later training.
Also noted is the yellow detection boxes are small compared to the size of the bus. It would be more ideal if the detection box is close to the actual size of the bus. This will probably require significant retraining.
It's been noticed that the detection of the busses is not close to 100%. While it is picking up a lot of busses, a large proportion are not being detected at all (the yellow box around them in debug screen). This can lead to false negatives, which is hard to detect without manual intervention.
Some of the cases are probably due to:
These will probably require retraining based on old footage, and/or on new footage.
Occasionally some false positives are noted due to people walking across road (and other causes not saved on video due to green light). These are in general more noticed due to the positive nature meaning that a video is saved and can be used for later training.
Also noted is the yellow detection boxes are small compared to the size of the bus. It would be more ideal if the detection box is close to the actual size of the bus. This will probably require significant retraining.