Hello, I am a beginner in deep learning and I want to build something like an adas (Advanced driver-assistance systems) using the Jetson Nano. Then DeepStream and Transfer Learning Toolkit sounded like a great option because I can use the pre-trained models and save some training time. I wanted to use TrafficCamNet or DashCamNet as my pre-trained model with my custom lane dataset to build an adas. However, searching for how to train models with TLT I recognized that Kitti annotation is the annotation method used by TLT and it is based on bounding boxes. Bounding boxes are not great to annotate lanes. I want a model with a lane, car and roadsigns detections, so my question is: how can I use the pre-trained models together with a lane detection technique in the same model?
TLT sounded great to me when I started my research, but now I am a little bit lost. If anyone could help me in any way I would be grateful. Thanks in advance.