Basic implementation of a Siamese network structure for detecting copyright infringement
A neural network is implemented here to detect copyrigt infringement violations in newly uploaded video files. The model takes the general form of a Siamese network, with two images filtered through the same convolutional neural network before a classification is made concerning the probability that the two images are a match. In order to train the model, frames were extracted from a set of high-resolution video files of movies and a corresponding set of lower-quality recordings of those movies. Randomly selected frames from the high-quality files were paired with the corresponding frame from the recorded version (a match) as well as a frame from another movie (not a match); these are referred to as triplets. Each of the images is run through the initial set of convolutional layers, which take the structure and weightings from the pre-trained VGG16 neural network, and a vector of length 4096 is returned. This serves as the input to a new set of top layers meant to classify the paired images as a match or not. In addition to the feature extraction and classification model, a means of aligning the recorded clip with the full movie is implemented to optimize the neural net performance. The top level function outputs a single probability of infringement that emerges as the average of the predicted match probabilities along the length of the potentially infringing clip.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size cimese_net-0.37-py2.7.egg (56.5 MB)||File type Egg||Python version 2.7||Upload date||Hashes View|
|Filename, size cimese_net-0.37.tar.gz (55.3 MB)||File type Source||Python version None||Upload date||Hashes View|