Plan of work

Phase	Description	Schedule
Create a dataset for training and testing	summary length > 2 hours resolutions: from 240p to 1080p train and test videos > 500 scene changes	June 7 - June 14
Test current solutions	Solution #1 Solution #2 do they make mistakes on the flashes? adjust the parameters to increase the f score	June 7 - June 21
Analyze the dependency of the current solution from the video resolution	test on downscaled videos on different resolutions: 240p, 360p, 480p, 720p, 1080p measure precision, recall, f score and speed on them make charts to analyze results	June 14 - June 28
Make adaptive threshold for the current solution	visualize results to analyze them estimate the gain	June 14 - June 28
Make metric for flash detection	mark up a few videos make flash detection metric based on histogram and mean intensity of the frame measure precision, recall, f score and speed	July 10 - first evaluation
New metrics	make a list of different scenechange implementation possible features motion vectors, edge detector, etc. dividing the frame into blocks and calculating metrics on them adaptive resolution scaling estimate gain and speed performance	July 10 - early August
Integration	Integrate the final solution into the product	until mid-August