meder411 / omnidepth-pytorch Goto Github PK

Hello, thanks for work. I tried to inference you network with rectnet weights and get strange result like this
. Maybe you know what can go wrong with that? And if you can, please give us some test panorama

Hello!

Thanks for your re-implement of the Omnidepth.

As you are the author of the Tangent-Images, how do you think about the OmniDepth's way to deal with the Spherical Convolution? Which way is better in your mind, the Tangent-Images or the RectNet?

Thanks.

Hello! Thanks for your reimplement of the Omnidepth. There is a confusion for me, that is, is the quantitave result and qualitative result performs well just as the author stated in the paper?

Or is there any difference? Can you show some result of your model?

Thx!

Hi, Meder. I think you may confused between io.imread() and cv2.imread(), the former one is adopted in your code, and Load in as RGB. While the latter, OpenCV would probably load BGR format that it need the code to meet the need for RGB. It just like the function defined in cv2.cvtColor(). Please let me know whether my comment is wrong or not. Thx!

I keep running into an error while building with the .yml file where I get :

TypeError: sequence item 0: expected str instance, Channel found

Has anyone else run into this? I've also attached a .txt of the error report below.

yml_setup_error.txt

Hello,

As mentioned in the paper of OmniDepth: Dense Depth Estimation forIndoors Spherical Panoramas. in the result section (quantitative result), inference result are deployed using new equirectangular images.

in this repositorty, test_omnidepth.py show performances using metrics. we would like to know if an inference script of the model exists or not.

Cordially

Dear meder411,

Thanks for your implementation on OmniDpeth that would be very beneficial to the community.

I was wondering if you will also upload your trained network, e.g. checkpoint_latest.pth, so we could have a solid reference when using your implementation. It also helps people who would like to try inference simply.

Best,
-- Luke

Hello,

I Downloaded the database mentioned in the readme, however to test the model I dont know how to define the parameters val_file_list and input_dir

I use Ubuntu 16.04 and Anaconda to build the environment. But I failed to run train_omnidepth.py file due to the error reported below:

/.local/lib/python3.7/site-packages/OpenEXR.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZTIN7Iex_2_27BaseExcE

By checking some blogs, I think the real problem may be that, the OpenEXR which is downloaded by pip may not suit to Python3.x. But I just can't find the way out.