KernelTransformerNetwork icon indicating copy to clipboard operation
KernelTransformerNetwork copied to clipboard

Published 20 hours ago • verified publisher icon sammy-su

Data and model download

Open tjinjin95 opened this issue 5 years ago • 11 comments

Hi, I am interested in your work about your KTN method. In your cfg.py file, I find the "DATA_DIR" and "MODEL_DIR" paths, could you give the download links in a Readme file. Thank you a lot. @sammy-su

tjinjin95 avatar Aug 06 '19 09:08 tjinjin95

Hi, did you get it? I need the download links, too.

Lishiyuan0813 avatar Dec 09 '19 12:12 Lishiyuan0813

Hi, please find the weights and models at our project webpage http://sammy-su.github.io/projects/ktn/

I will update the Readme later.

sammy-su avatar Dec 10 '19 06:12 sammy-su

Hi, did you figure out how to get data in the "DATA_DIR" path? thank you!

ghost avatar Dec 25 '19 10:12 ghost

The structure of DATA_DIR is DATA_DIR train.txt valid.txt pixel/ imagenet1_1/ ... {src_cnn}{layer}/

Each line in train.txt and valid.txt contains an image name, and the loader will try to read pixel/{image_name}.jpg from the image and {src_cnn}{layer}/{image_name}.h5 for the intermediate output.

Please let me know if you need the data we used in our experiment. There's no ready to download version currently because of the data size.

sammy-su avatar Dec 25 '19 20:12 sammy-su

Hi! Thanks for your quick reply! But, still have some questions:

  1. Are the images in the 'pixel' directory equirectangular format?
  2. Are the h5 file generated by doing back projection on each of the row pixel of the equirectangular and then feed those images into source CNN as what your previous work(i.e. generate_sources.py in Spherical-Convolution) done?

Thank you so much!

ghost avatar Dec 26 '19 14:12 ghost

  1. Yes, the images are in equirectangular projection with 640 x 320 resolution.

  2. Yes, we use exactly the same data in Spherical-Convolution and KernelTransformerNetwork.

sammy-su avatar Dec 27 '19 03:12 sammy-su

Will the used datasets in the paper be released? I'd like to reproduce the results in the paper.

Nicholasli1995 avatar Jan 09 '20 05:01 Nicholasli1995

Please try the new link at http://vision.cs.utexas.edu/projects/ktn/ktn.tar.gz

On Wed, Jan 8, 2020 at 10:32 PM tjinjin95 [email protected] wrote:

[image: image] https://user-images.githubusercontent.com/30750867/72043604-90216600-32ec-11ea-954f-9ec68403a1d3.png Hi, would you provide other download address, it is too difficult to download this file. Thanks a lot.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/sammy-su/KernelTransformerNetwork/issues/1?email_source=notifications&email_token=AASI6IZ2ZBTGQ4W3IWJ2KQ3Q43AHBA5CNFSM4IJUDZF2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEIPFCNA#issuecomment-572412212, or unsubscribe https://github.com/notifications/unsubscribe-auth/AASI6IZJPWZ2XFJQAF247RDQ43AHBANCNFSM4IJUDZFQ .

sammy-su avatar Jan 13 '20 04:01 sammy-su

Hi, I want download the data of "DATA_DIR". I would be very grateful if I could receive your reply. @sammy-su

tjinjin95 avatar Feb 02 '20 10:02 tjinjin95

Hi, I am interested in your work, is there a link available to download data of "DATA_DIR"? @sammy-su
. I'd like to reproduce the results in the paper, and try for other architectures too. I would be very grateful.

Artcs1 avatar May 16 '20 13:05 Artcs1

Because the total data size is about 4TB, it is hard to share the data under DATA_DIR. The data is generated using the script in Spherical-Convolution. Please refer to the following script to generate the data. https://github.com/sammy-su/Spherical-Convolution/blob/master/bin/generate_sources.py

sammy-su avatar Aug 05 '20 05:08 sammy-su