how to train?

[Chenhait]

I try to run 'python train.py',but always 0%

[ynjiun]

using kitti dataset as an example: to run training:

python train.py args_train_kitti_eigen.txt

Potential issues for "always 0%"

1. configure batch size in args_train_kitti_eigen.txt: The batch size default is 16 (--bs 16). If your GPU only has 11GB such as 1080Ti, then I would recommend to set it to 3 (--bs 3).
2. file listed in train_test_inputs/kitti_eigen_train_files_with_gt.txt may not exist in your dataset: There might be some files listed in this text file do not exist in your dataset. You might need to write a small script to check and remove those filenames not exist. Likewise, you might need to check kitti_eigen_test_files_with_gt.txt as well.

[Chenhait]

thank you so much for your reply!!! i indeed meet the problem for "always 0%" when i try to train. i think maybe the second way can solve my problem. however, i am a beginner of deep learning, and i have no idea about the meaning of kitti_eigen_train_files_with_gt.txt. i do not find where to find the 2011_09_26_drive_0001_sync/proj_depth/groundtruth/image_00. thank you if any advance

[ynjiun]

@Chenhait

you may need to download the raw images and ground truth depth images from kitti for raw images at here for depth images at here you might need to sign up an acount at kitti to download the dataset once downloaded, you may need to set it up to the directory as prescribed in args_train_kitti_eigen.txt --data_path put your raw images directory here --gt_path put your depth images directory here likewise need to modify the following two paths as well --data_path_eval put your raw images directory here --gt_path_eval your depth images directory here Well at this point you might be able to train your model, however, if it got stuck again, that might imply your path did not set it correctly or as previous #2 stated such that file not found, or there might be missing some files in your dataset so need to modify according to #2 suggested.

[Chenhait]

@Chenhait

you may need to download the raw images and ground truth depth images from kitti for raw images at here for depth images at here you might need to sign up an acount at kitti to download the dataset once downloaded, you may need to set it up to the directory as prescribed in args_train_kitti_eigen.txt --data_path put your raw images directory here --gt_path put your depth images directory here likewise need to modify the following two paths as well --data_path_eval put your raw images directory here --gt_path_eval your depth images directory here Well at this point you might be able to train your model, however, if it got stuck again, that might imply your path did not set it correctly or as previous #2 stated such that file not found, or there might be missing some files in your dataset so need to modify according to #2 suggested.

thank you so much for your kindness!!! you like the light of my life

[agenthong]

@Chenhait you may need to download the raw images and ground truth depth images from kitti for raw images at here for depth images at here you might need to sign up an acount at kitti to download the dataset once downloaded, you may need to set it up to the directory as prescribed in args_train_kitti_eigen.txt --data_path put your raw images directory here --gt_path put your depth images directory here likewise need to modify the following two paths as well --data_path_eval put your raw images directory here --gt_path_eval your depth images directory here Well at this point you might be able to train your model, however, if it got stuck again, that might imply your path did not set it correctly or as previous #2 stated such that file not found, or there might be missing some files in your dataset so need to modify according to #2 suggested.

thank you so much for your kindness!!! you like the light of my life

兄弟，有点搞笑

[Chenhait]

@Chenhait you may need to download the raw images and ground truth depth images from kitti for raw images at here for depth images at here you might need to sign up an acount at kitti to download the dataset once downloaded, you may need to set it up to the directory as prescribed in args_train_kitti_eigen.txt --data_path put your raw images directory here --gt_path put your depth images directory here likewise need to modify the following two paths as well --data_path_eval put your raw images directory here --gt_path_eval your depth images directory here Well at this point you might be able to train your model, however, if it got stuck again, that might imply your path did not set it correctly or as previous #2 stated such that file not found, or there might be missing some files in your dataset so need to modify according to #2 suggested.

thank you so much for your kindness!!! you like the light of my life

兄弟，有点搞笑 希望多欢乐，少烦恼

[Chenhait]

@Chenhait

you may need to download the raw images and ground truth depth images from kitti for raw images at here for depth images at here you might need to sign up an acount at kitti to download the dataset once downloaded, you may need to set it up to the directory as prescribed in args_train_kitti_eigen.txt --data_path put your raw images directory here --gt_path put your depth images directory here likewise need to modify the following two paths as well --data_path_eval put your raw images directory here --gt_path_eval your depth images directory here Well at this point you might be able to train your model, however, if it got stuck again, that might imply your path did not set it correctly or as previous #2 stated such that file not found, or there might be missing some files in your dataset so need to modify according to #2 suggested.

sorry to bather you again. since i am too not smart. i download a part of data, and try to train. i change the --bs=3, and set the directory, i change the list in train_test_inputs/kitti_eigen_train_files_with_gt.txt（i am not sure this step is right）. however , it still can not run. it would be very appreciate if any advance!!! hope you happy everyday.

[liuxu0303]

@Chenhait you may need to download the raw images and ground truth depth images from kitti for raw images at here for depth images at here you might need to sign up an acount at kitti to download the dataset once downloaded, you may need to set it up to the directory as prescribed in args_train_kitti_eigen.txt --data_path put your raw images directory here --gt_path put your depth images directory here likewise need to modify the following two paths as well --data_path_eval put your raw images directory here --gt_path_eval your depth images directory here Well at this point you might be able to train your model, however, if it got stuck again, that might imply your path did not set it correctly or as previous #2 stated such that file not found, or there might be missing some files in your dataset so need to modify according to #2 suggested.

sorry to bather you again. since i am too not smart. i download a part of data, and try to train. i change the --bs=3, and set the directory, i change the list in train_test_inputs/kitti_eigen_train_files_with_gt.txt（i am not sure this step is right）. however , it still can not run. it would be very appreciate if any advance!!! hope you happy everyday.

请问你解决了吗？我也遇到了同样的问题，一直0%

[Chenhait]

我也遇到了同样的问题

我的可以跑了，不过我对kitti数据有点不熟，如果你愿意和我讨论的话，加我一下908616145

[JK1201]

我也遇到了同样的问题

我的可以跑了，不过我对kitti数据有点不熟，如果你愿意和我讨论的话，加我一下908616145

你好，我遇到了你在pytorch3d下面提出的问题 from pytorch3d import _C ImportError: libc10_cuda.so: cannot open shared object file: No such file or directory 请问你最后是怎么解决的

[mujinsong]

我也遇到了同样的问题

我的可以跑了，不过我对kitti数据有点不熟，如果你愿意和我讨论的话，加我一下908616145

你好，我遇到了你在pytorch3d下面提出的问题 from pytorch3d import _C ImportError: libc10_cuda.so: cannot open shared object file: No such file or directory 请问你最后是怎么解决的

你的版本并不对应，pytorch3d的版本应该和torch版本对应

[9796l]

你好，我想请问一下，我在进行实验的时候总是会报错，有的时候会报通道的错误，由于我使用的不是kitti和nyu，想请问一下，这个实验当中使用的深度图的shape，是1通道的灰度图吗，还是3通道的？非常期待您的回答，非常感谢。

[9796l]

你好，我想请问一下，我在进行实验的时候总是会报错，有的时候会报通道的错误，由于我使用的不是kitti和nyu，想请问一下，这个实验当中使用的深度图的shape，是1通道的灰度图吗，还是3通道的？非常期待您的回答，非常感谢。

[mujinsong]

你好，我想请问一下，我在进行实验的时候总是会报错，有的时候会报通道的错误，由于我使用的不是kitti和nyu，想请问一下，这个实验当中使用的深度图的shape，是1通道的灰度图吗，还是3通道的？非常期待您的回答，非常感谢。

我不知道你报的错是什么,图应该是三通道吧(我看nyu里面的是彩色),你可以把nyu的数据下载下来(应该是一个2.8G的.mat文件,但你可以解析出来),然后看看里面的数据,程序是对像素有要求的,nyu要求是640*480,否则报错

[9796l]

好的，非常感谢你的解答，方便加一下联系方式交流吗？Q1794753389，非常感谢