Closed alikureishy closed 5 years ago
I suggest pretrained SSD_GRAPH or FASTER_RCNN. In this case, we have only to fine-tune the existing model.
Without any fine-tuning I was able to detect traffic light contours on the image in most cases.
The most elegant method is to fine-tune the model, and the CNN will classifier the traffic lights's color too.
This file will expose a method:
def create_model_for_training(.....):
... and return the model architecture as below:
This will be used by the tl_trainer.py node. Having this as a separate file allows the architecture to be implemented independently from the trainer.