I wonder if we need to add some constraints like limiting the output values from -1 to +1 and then calculate the loss? Or some other constraints (e.g. loss) to make the sum of the square of sin and cos value equal to 1. I have already found that when visualizing, we limit the output values from -1 to +1.
What is the performance of the current version? Is the direction accurate?
I am not very familiar with TensorFlow and may miss some details. Hope you can give me some guidelines.
Dear songtao, Thanks for your great work.
I have a question about the direction head and loss.
I find that https://github.com/songtaohe/LaneExtraction/blob/master/code/cnnmodels/resnet34unet.py#L89-L95 calculates the direction head and just uses an FFN without any constraints.
I wonder if we need to add some constraints like limiting the output values from -1 to +1 and then calculate the loss? Or some other constraints (e.g. loss) to make the sum of the square of sin and cos value equal to 1. I have already found that when visualizing, we limit the output values from -1 to +1.
What is the performance of the current version? Is the direction accurate?
I am not very familiar with TensorFlow and may miss some details. Hope you can give me some guidelines.