Feature extraction

[Veldhoen]

Apply code by Teng:

Input: keyframes + audio spectograms
Output: binarized feature matrix

• Preprocess keyframes & spectograms
• Load in pretrained model from model files
• Apply pretrained model on data
• Binarize resulting feature matrix (C++ rather than Python). Teng will provide an example matrix for testing this bit in isolation.
• Use GPU for processing

[Veldhoen]

First steps have been implemented (in a rudimentary way).
Applying the model fails on dimensionality issues. Teng is now looking into the dimensionality of the representations that come out of worker 1, that do not seem to match the expectations.

[Veldhoen]

Binarization moved to separate issue, nice to have

[Veldhoen]

GPU moved to separate issue, requires installation etc.