OpenCV's incompatibility about c3d-keras HOT 2 CLOSED

Although, when I use opencv-python, I'm able to replicate the results.
For my video dataset, I'm using the above ffmpeg command to burst and save frames to feed into the input pipeline of tensorflow. On the other hand, is I use opencv then saving numpy object for bursted frames is a disadvantage in terms of storage capacity.

If possible, can you share how you go about it ?

P.S. the histograms for video clip for opencv v/s specific ffmpeg command

Shape of vid opencv --> (16, 128, 171, 3)
(array([446550, 141813,  80391,  75870,  48795,  77991, 131775,  37860,
         8763,    816]), array([   0. ,   24.6,   49.2,   73.8,   98.4,  123. ,  147.6,  172.2,
        196.8,  221.4,  246. ]))

Shape of vid using the above ffmpeg --> (16, 128, 171, 3)
(array([428868, 149859,  87015,  75621,  57087,  86070, 127770,  32820,
         4953,    561]), array([   0.,   25.,   50.,   75.,  100.,  125.,  150.,  175.,  200.,
        225.,  250.]))

from c3d-keras.

Hope you figured out the way to deal with opencv. SInce this is not directly related to Keras implementation of C3D, I'm closing this issue.

from c3d-keras.