Python script to automate the creation of Dota 2 compelation videos from full games. This will first download the full video, gather the relevant timestamps of clips, split the video up at these timestamps, add the dotaClipz logo, automatically create a thumbnail, then (future update) upload the video to youtube using the Data API.
The timestamps are gathered from the video by parsing frames of the video, using a modified binary searchish itteration for efficiency, and uses a custom CNN to read the scoreboard of each frame. When the score changes, we create a clip around it.
This CNN was built similarly to a top solution in predicting the MNIST dataset, a common problem used in acedamia. The pixelated digits are between 5-8x10 pixels which is very similar to the 12x12 pixel digits used in the MNIST handwritten digit dataset. It predicts the correct digit at an accuracy of 99.96%
Using moviepy, we split that full video along the timestamps, add the dotaClips logo, and combine the clips into one full video.
The thumbnail creation is done by scraping the team logos from Liquipedia and pasting them ontop of a template created by @Allie_Leto
Eventually we will use the youtube data API to upload these compelations automatically. We can only upload 6 videos a day.
1: Since the projects that enable the YouTube Data API have a default quota allocation of 10,000 units per day [2] and a video upload has a cost of approximately 1,600 units [3]: 10,000 / 1,600 = 6.25.
The propper version of FFMPEG for ur system needs to be installed and the exe needs to either be in /bin or your path. For me this was ffmpeg-release-arm64-static.tar.xz
Need 5 folders in base dir, "finishedVideos", "logs", "thumbnailCreation", "videoClips", "rawVideos"