June 21 - June 28 Weekly Status

This week I tested the dependency of speed on resolution of the video. So far, results seem strange because downsampling from 4k to 2k didn’t affect much speed of the algorithm (approximately 1.2-1.3x faster).
Then I measured the time complexity of downsampling alone. Inside the video frame reading function, I set the loop for 10000 iterations - at each of them the memory was allocated for a new frame of smaller size and the planes to which downsampling was applied were copied to the new frame valuable. It turned out to be about 200 iterations per second, which is way faster than 4k videos regular scenechange detector processing speed (2.5 frames per second). So the downsampling itself shouldn't affect the speed much. So, here's the problem to solve.
I made an output of the metric of the algorithm to the json file and visualized it in a few charts. I made a blog post about it. Based on visualization I made some conclusions for future work, which are written at the end of the post.