Closed arianaa30 closed 5 years ago
Thanks! Yes, to preserve the original resolution you can specify --mov_width_px -1
(or --jpg_width_px -1
or --gif_width_px -1
depending on what your output format is). When the value is negative, it will copy the input resolution. See these lines in include/hecate/hecate.hpp
and these lines in src/hecate/hetace_main.cpp
for more information.
Great! So right now the selection is based on "dynamism" score, right? Do you have any recent work or consider improving based on deep models to score "interestingness" or something?
Some people have already tried incorporating deep nets to replace the scoring mechanism, reporting improved performance (I don't think they published a paper, but I know that version is deployed in production). I personally don't have an immediate plan to do so, unfortunately.
Interesting work. I compiled and was trying the video summarization, but noticed it reduces the output resolution/quality significantly. Any ways we can preserve the input resolution?