config about DeepLO

[rginjapan]

DeepLIO Network

deeplio: dropout: 0.25 pretrained: false model-path: "" lidar-feat-net: name: "lidar-feat-simple-1" pretrained: false model-path: "" imu-feat-net: name: "imu-feat-rnn" # See the name is commented out pretrained: false model-path: "" odom-feat-net: name: # "odom-feat-fc" # See the name is commented out pretrained: false model-path: "" fusion-net: "fusion-layer"

why you comment odom-feat-fc, I think odom is still necessary but fusion-net should be commented.

[rginjapan]

I have trained lidarsegpoint net + odometry RNN net, the result is very bad no matter in training dataset and test dataset. Could you give me some instructions about how to reproduce the result of DeepLO in the paper? Btw, in DeepVO (https://github.com/ChiWeiHsiao/DeepVO-pytorch), the network is CNN + LSTM, which I think is similar with lidarsegpoint net + odometry RNN net I have tested. But the results are very bad in my test.

[rginjapan]

I know why you did not comment fusion-net in DeepLO, because the fusion-net will be created only when both lidar-net and imu-net existed. So why you comment out odom-net in DeepLO?