Nearly identical to #27 however this task specifically looks at complex models built atop either post-graph features or the raw images. For example, an out-of-the-box ResNet-50 would be a good baseline for the raw images, while a more complex LSTM-MLP network may be a good place to start for the post-graph features. This should demonstrate whether such a representation can generate more useful features than a traditional convolution based representation.
Nearly identical to #27 however this task specifically looks at complex models built atop either post-graph features or the raw images. For example, an out-of-the-box ResNet-50 would be a good baseline for the raw images, while a more complex LSTM-MLP network may be a good place to start for the post-graph features. This should demonstrate whether such a representation can generate more useful features than a traditional convolution based representation.