vasanthkalingeri / CommercialDetection

GSOC 2015 project for Red Hen Labs
86 stars 36 forks source link

Does not detect scene changes #2

Open EvanWeiner opened 8 years ago

EvanWeiner commented 8 years ago

When attempting to detect commercials (e.g. run with -r flag) there appears to be no automatic scene change detection. My output.txt just says the entire file is unclassified. There appears to be code to perform this auto detection in the DetectSilence class, but it is not referenced anywhere in the rest of the project. It seems that this project can only detect previousy labeled commercials.

Can you please show a simple "hello world" example that can demonstrate the automatic detection of commercials without any prior labels? Starting from scratch. I have a video file and want to get the timings for where possible commercials may be located. Thank you.

vasanthkalingeri commented 8 years ago

That was an old portion of the code which can be integrated if required, but is not. The main aim is to have 100% recognition on known commercials so audio fingerprinting is used to detect commercials. There is no necessity to perform scene change detection (whose accuracy isn't 100%). This is an optional part to extend the project to perform automatic commercial detection.

You should have a database of known commercials if you want this to work on your video file. The program detects those commercials and provides a web interface to add more commercials to the database through manual tagging. It does not perform automatic commercial detection yet.

EvanWeiner commented 8 years ago

Thank you Vasanth for your fast response. I am very interested in your "Finding scene changes" section in your writeup here "vasanthkalingeri.github.io/CommercialDetection." I'd like to hear your advice on how to extend the project to perform this automatic commercial detection with the DetectSilence class or others. I also found in your writeup you cited this branch (https://github.com/vasanthkalingeri/CommercialDetection/tree/094a883c3293bd9c406e9e7ec44a2db292ac4148) -- can you please advice which code parts are needed for this effort?