search

ChiWeiHsiao / DeepVO-pytorch

PyTorch Implementation of DeepVO
358 stars 106 forks source link

train problem #21

Open ujgygy656565 opened 5 years ago

ujgygy656565 commented 5 years ago

Hello, I am a novice. I have two questions. I set resume to false. I want to train it from the beginning. But when I trained 100 to 200 epochs, the results of the tests and visualize are very poor. I haven't changed your parameters and code. What's the reason? The second question is, the img-means and stds in params are calculated by preprocess? Thank you very much.

ujgygy656565 commented 5 years ago

I set resume to false. According to the parameters given in params, I trained to 100 epoch and 250 epoch, and tested them separately. But the effect of visualize is very poor. Is epoch not enough or where can be adjusted? Can you teach me?

ujgygy656565 commented 5 years ago

Why don't I change the parameters and train from scratch to 400 epoch and the result is like this. Thank you. image image image

alexart13 commented 5 years ago

@ujgygy656565 Did you use a pretrained FlowNet?

ujgygy656565 commented 5 years ago

@ujgygy656565 Did you use a pretrained FlowNet?

yes，I changed pretained flownet in params to. / pretrained / flownets_bn_EPE2.459. pth. tar. What may be my problem? Is the label relative position and posture in training? In which file is it expressed? Thank you.

alexart13 commented 5 years ago

try flownets_EPE1.951.pth.tar. Also it seems that your NN is undertrained.

ujgygy656565 commented 5 years ago

try flownets_EPE1.951.pth.tar. Also it seems that your NN is undertrained.

Thank you. I've got the normal trajectory now. How to calculate the evaluation index of KITTI SLAM?

DexterFixxor commented 4 years ago

@ujgygy656565 Hello, what did you do to solve your problem? It appears that I have similar situation. While training, my RMSE is very small for both train and valid set, but visualization is just bad.

la735681253 commented 4 years ago

@ujgygy656565 Hello, what did you do to solve your problem? It appears that I have similar situation. While training, my RMSE is very small for both train and valid set, but visualization is just bad.

hi，when i have download the flownets_EPE1.951.pth.tar，it shown that The compressed file format is unknown or the data is corrupted，what is the problem？

ghost commented 4 years ago

try flownets_EPE1.951.pth.tar. Also it seems that your NN is undertrained.

Thank you. I've got the normal trajectory now. How to calculate the evaluation index of KITTI SLAM?

hello, Have you solved the problem? I also don't know how to calculate the evaluation index of KITTI SLAM