This is another fun project that allows automatically downloading youtube videos and running semantic segmentation on the videos. It may be used to quickly test/visualize pretrained models on the desired youtube videos.
An example is:
Install the requirements by running the code below at the command window, at the project's directory (if pytorch doesn't install, install from PyTorch):
pip install -r requirements.txt
At the examples given in jupyter notebooks, you need to replace with the youtube link that you would like to run the semantic segmentation algorithm:
- example_1: - With the given youtube link, it downloads the video, separates into a set of images (frame rate is adjustable), runs the desired semantic segmentation algorithm and reconstructs a video as the original video at the background overlayed with the object detection from the net.
The project requires only the youtube link at minimum, yet it is easily costumizeble with a few lines of code. For example, you can pass a different net as:
process_and_run(net='vgg11').run(youtube_link)
You might need to modify the semantic_images.py according to the: PyTorch pretrained models for that specific net.
There are further modifications can be done easily, such as the frame rate limitation when separating the video into images. "frame_rate_limiter" in the separate_to_jpgs.py can be decreased to populate more images. If the program takes too much time, consider decreasing that value.