Jia-Ren Chang comments

Results 98 comments of


                                            Jia-Ren Chang

m.weight.data.normal_(0, math.sqrt(2. / n)) what does it mean?

@zhang405744522 It is weight initialization. Please refer to [Kaiming He 2015 paper](https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/He_Delving_Deep_into_ICCV_2015_paper.pdf)

Test the pre-trained model with EuRoc Vicon Room

@Zchappie Hi. Please notice your torchvision version, you should use 0.2.0.

Test the pre-trained model with EuRoc Vicon Room

@Zchappie Hi, would you provide me a pair of images to test?

Test the pre-trained model with EuRoc Vicon Room

Hi, @Zchappie I found that this pair of images is not correctly rectified. ![test](https://user-images.githubusercontent.com/11732099/60525478-fff0db80-9d20-11e9-88e9-d5240fc419ba.png)

Test the pre-trained model with EuRoc Vicon Room

@Zchappie Here is my result of your image pair. I used trained KITTI2015 model. ![test_color](https://user-images.githubusercontent.com/11732099/60888242-ebfd2a80-a288-11e9-9204-47159795b90a.png)

Test the pre-trained model with EuRoc Vicon Room

Hi, @Zchappie I didn't tune any parameters. I used the KITTI 2015 pre-trained model. You can find the download link on Readme. The only difference is that I loaded the...

Test the pre-trained model with EuRoc Vicon Room

@Zchappie I did not really catch what you means. My system is ubuntu 16.04, and an NVIDIA 1080 ti for inference.

Test the pre-trained model with EuRoc Vicon Room

@Zchappie Yes, it would be better to have a GPU to infer. BTW, a GPU with 6~8GB memory is a good choice.

Test the pre-trained model with EuRoc Vicon Room

@Zchappie It may due to the version of skimage. I am not sure. My skimage version is 0.13.0. scikit-image 0.13.0

About submission on the kitti dataset

@DrugRui Yes, only KITTI 2015 dataset for training the model of KITTI 2015 submission.