DeepFusion network structure

[faikit]

Hi,

we are trying to reproduce the result, but quite a lot details are missing of the DeepFusion netwrok. Can you describe some details of the network? Like in which stages fuse the point cloud and image, and etc.

[LiYingwei]

We just released the code. You could start to read the code from here: https://github.com/tensorflow/lingvo/blob/master/lingvo/tasks/car/params/waymo_deepfusion.py#L797

You could look at MultiModalFeaturizer to know the fuse stage:

lingvo/lingvo/tasks/car/deep_fusion.py

Line 76 in 6b75b05

class MultiModalFeaturizer(base_layer.BaseLayer):