jiupinjia / skyar Goto Github PK

Hello!

First of all, great work!
I was wondering if you can explain a little bit the meaning of concat for imgtile and how it works.

Thank you!

Running through the Google Colab notebook, Wandb only seems to contain GIF outputs for my videos. The notebook says I can download a full video rather than the preview at /SkyAR/demo.avi but I see no AVI files in my Wandb media. What am I missing here?

Great job and thank you.
I set up an environment to run the demo correctly. But when I process a 720x1280 video, the picture or video as the sky is always stretched vertically.
Is it because this program cannot handle other aspect ratio videos, or do I need to change any parameters?

I I will be grateful for any suggestions from you.

Thanks.

I have a problem while runniing the project. there is a ./torch folder under my c drive.but my other programs are under F,i dont know if it has anything with it. :(

Awesome work. I like this project very much. So I am implementing this project on another deep learning framework (paddlepaddle) to make it easier to use. Now the Demo has been Done. Online trying link emm, this page is written in Chinese. The github code will be release soon.

And I have a small question that can I upload my implementation to paddlehub (a open source pre-training model management tool similar to the Pytorch Hub)? I want more people to simply use this awesome project.

Hello, I am having trouble running train.py

(base) [contrast@cosmos SkyAR-main]$ python train.py \

--dataset cvprw2020-ade20K-defg
--checkpoint_dir checkpoints
--vis_dir val_out
--in_size 384
--max_num_epochs 200
--lr 1e-4
--batch_size 8
--net_G coord_resnet50
Traceback (most recent call last):
File "train.py", line 4, in
import utils
File "/home/contrast/Desktop/SkyAR-main/utils.py", line 4, in
from skimage.measure import compare_ssim as sk_cpt_ssim
ImportError: cannot import name 'compare_ssim' from 'skimage.measure' (/home/contrast/Applications/mc/lib/python3.8/site-packages/skimage/measure/init.py)
(base) [contrast@cosmos SkyAR-main]$

I tried a first test video and the tracking seems not to work on a simple roadvideo.
https://www.youtube.com/watch?v=vy38G5g7FaY

I receive a lot of no good matching points.

Can I adjust the tracking?

I tried to adjust the crop, but it doesn't work. Thanks for sharing this project.

Hello! I have a problem with installing Requirements.txt. I can't correctly install modules and every time gets an error.

• scipy send No lapack/blas resources found. and other libraries. Can you send a list of all commands for pretty installing?
  To install all modules I'm using pip install Requirements.txt.

can you please add a google colab for inference thanks

After processing the image and videos, there is no mp4 in the "jpg_output".
So , how to save the result.
please

Hello,

First of all, nice job. I was wondering why did you only encode the y coordinate for the CoordConv instead of both x, y or only x. Can you explain the reason behind this?

Thank you!

from skimage.measure import compare_ssim as sk_cpt_ssim in utils.py
import problems: ImportError: cannot import name 'compare_ssim' from 'skimage.measure' (D:\anaconda\envs\pytorch\lib\site-packages\skimage\measure_init_.py)
I find that the latest version of scikit-image doesn't have "compare_ssim".
Maybe an update in requirements.txt is needed

I can see a real use case for this is my realestate videography to replace the cloudy sky of an overcast day to a sunny sky, however after testing with some cloudy footage in google collab it seems like it doesn't work with overcast skies. It just fails to process most of the frames. Or am i doing something wrong?

Hi. First, thanks for providing this in github for us to experiment.

I've tried to use a custom video, but it seems like the skybox is always following the camera.

This is an example:
https://youtu.be/Ogt2GM-H2LU

And this is my json file:

{
  "net_G": "coord_resnet50",
  "ckptdir": "./checkpoints_G_coord_resnet50",

  "input_mode": "video",
  "datadir": "./test_videos/test.mp4",
  "skybox": "district9ship.jpg",

  "in_size_w": 384,
  "in_size_h": 384,
  "out_size_w": 1280,
  "out_size_h": 720,

  "skybox_center_crop": 0.6,
  "auto_light_matching": false,
  "relighting_factor": 0.9,
  "recoloring_factor": 1.0,
  "halo_effect": true,

  "output_dir": "./jpg_output",
  "save_jpgs": true
}

The file is 1280x720.

Any idea why this happens?

Regards.

Bro, I've been playing with your library all evening!
Amazing work! Made my friends worry that I am on another planet)))
I wrote a small tutorial if anyone is interested.

Shure, this work is a good case when there is something to show in a code presentation in contrast with a usual boring code description.

Hello, can you suggest where in the code I can make changes to work on images only instead of videos?

Hey! Great library!
I'm concerned about the noncommercial limitation.
I want you to consider changing this and here's why:

1. It only limits the project and where people can go with it.
   • It means the project can't be included in a commercial work,
     no matter how greatly that commercial work may add to it.
   • It's not what's keeping your code free, as your code is here on github for everybody to see.
2. It only complicates things when people use your code.
   • Someone may include it in a GPL-licensed repo
     and then someone may use that GPL-licensed repo commercially,
     like how linux distros can come with a system,
     and even if they release the entire system for people to see,
     they are still using it commercially by selling a system with it on it.
     This could lead to accidental misuse of your code under this license.
3. Maybe a different license wouldn't be so bad. :)
   • It seems like you would actually prefer the GPL license.
     it's what Linux uses and everyone all knows that Linux is pretty much the mother of all free software.
   • Or you could consider:
     Apache 2.0,
     MPL,
     etc
   • If you want it to be truly free, why not just
     MIT,
     CC0,
     The Unlicense,
     etc, etc.
     There are plenty of great licensing options that don't restrict the software to this extent.
     I think you would find that more people would be willing to use this software,
     even noncommercially, if you changed this. :)

Thank you SO MUCH for creating this software and for considering this request.
I appreciate it greatly! 👍