search

IRMVLab / EfficientLO-Net

EfficientLO-Net: Efficient 3D Deep LiDAR Odometry (PAMI 2022)
70 stars 4 forks source link

About pytorch version code #2

Open Yeah2333 opened 1 year ago

Yeah2333 commented 1 year ago

@XINRUII When will it be released？

XINRUII commented 1 year ago

In about a month or so，thx

Yeah2333 commented 1 year ago

@XINRUII I am very much looking forward to the code of the pytorch version. By the way, I train efficient lo-net using 1080ti and Tensorflow 1.12.0. The result is worse than that in the paper. May cause by low batch_size(2) because of low GPU memory(1080ti - 11g). And TensorFlow 1 didn't adapt 3090 because of cuda version.

JoaoAMarinho commented 7 months ago

Any update on this release?

wennycooper commented 1 month ago

@XINRUII I am very much looking forward to the code of the pytorch version. By the way, I train efficient lo-net using 1080ti and Tensorflow 1.12.0. The result is worse than that in the paper. May cause by low batch_size(2) because of low GPU memory(1080ti - 11g). And TensorFlow 1 didn't adapt 3090 because of cuda version.

Hi, did you succeed to trained the model?

I think the repo has many mistakes. For example, it missed tools folder for running kitti_evaluation.py and in https://github.com/IRMVLab/EfficientLO-Net/blob/master/main.py#L224 It calls the train_one_epoch as below:

train_one_epoch(sess, ops, train_writer, train_list = TRAIN_LIST)

But the function definition of train_one_epoch is different: in https://github.com/IRMVLab/EfficientLO-Net/blob/master/main.py#L344

def train_one_epoch(sess, ops, train_writer):