search

BichenWuUCB / SqueezeSeg

Implementation of SqueezeSeg, convolutional neural networks for LiDAR point clout segmentation
BSD 2-Clause "Simplified" License
561 stars 239 forks source link

The problem in result by eval in val data #12

Open ywangeq opened 6 years ago

ywangeq commented 6 years ago

Hi, when I follow the code and do the evaluation, the result quite different with the result which shown in the paper, especially in pedestrian and cyclist. And I do not change any parameters. Is this reasonable?

stratomaster31 commented 6 years ago

Have you previouslly trained the model and saved any checkpoint? Check if you have changed the path to the checkpoints directory in "eval.py". If no checkpoint exists, I think the evaluation runs with random weights (not-trainend) and the results will be quite different, I guess very low... It is only an idea

stratomaster31 commented 6 years ago

Another issue is that Table 1 in the paper does not specify which dataset was evaluated (training or evaluation). It should be evaluation, but it is no clear...

arunaswamy1985 commented 5 years ago

Hi , what is the training dataset set used in kitti , is it possible to retrain the network for 32 layer data ?