As the time of this writing (February, 2021) FGVC is one of the best video inpainting projects out there.
The authors did a great work and made it as easy a possible for others to use their video inpaiting tool. They even made a google colab notebook.
However, their tool is command-line based. I struggled a little to use it with my own data in colab. I was not very confortable with the workflow.
- Every time I needed to run it, I needed to clone the tool and download the models again.
- It was hard for me to upload my own data and adjust the tool parameters using the %%bash magic command.
- It was hard for me to se the results after running the video completion.
So I decided to adapt the original google colab notebook for my personal use.
I think it might me useful for others. Check it in nbviewer or in google colab.
- I wrote a function to view data data (folders with image files) as animated gif. It allows to easily see the results in google colab.
- The FGVC tool and the deep learning models are saved to google drive, which is mounted at the beginning of the notebook.
- Both input data and results are also saved to google drive.
- The FGVC tool now runs in Python instead of bash.
- To do that I change the current working directory in Python, import the video_completion_seamless() function from video_completion.py
- The I simulate the argparse result with a custom dict and call the function
- I also added a little bit of code to measure the execution time of video_completion_seamless() function.
- Open FGVC_video_completion.ipynb in google colab.
- Mount your google drive.
- (just once) Download the tool and model to your google drive.
- You can use the commented bash code block, located in cell 2.1. Run it once and comment it again.
- As february 2021, there is no need to install any package on google colab (pytorch, etc).
- Activate GPU on colab.
- Configure the input data at cell 3.3.
- By default it uses the Tennis demo from FGVC.
- Run all cells.
- See the results at cell 3.6.