zhangfuyang/rl_CARLA

What do you use as input for the network?

qiaoxianfeng opened this issue · 2 comments

qiaoxianfeng commented

I see that you are not directly using RGB images as input, so what do you use as input to the network?

zhangfuyang commented

I used a pre-trained network to encode feature vector. which is pretrained by imitation learning.

There is the code for computing the feature. https://github.com/zhangfuyang/rl_CARLA/blob/master/environment/imitation/imitation_learning.py#L62

qiaoxianfeng commented

Can you explain in detail what some feature vectors refer to? Is the street and vehicle information extracted from the RGB image?