Multimodal Representation of Advertisements Using Segment-level Autoencoders

This repository contains scripts to replicate the paper titled, "Multimodal Representation of Advertisements Using Segment-level Autoencoders" published at the 20th ACM International Conference on Multimodal Interaction, 2018. The paper can be found here

The following python 2.7 packages need to be installed in order to run the scripts

scipy==1.0.0
tqdm==4.11.2
numpy==1.11.0
pandas==0.22.0
six==1.10.0
tensorflow_gpu==1.4.0
joblib==0.11
Keras==2.1.5
scikit_learn==0.19.2
tensorflow==1.10.1

Brief description of the scripts

1. feature_extraction_scripts: C3D activity-net and vvgish features
2. architectures.py: model configs for all experiments in the paper
3. classification_experiments.py: classify using the embeddings
4. create_baseline_experiments: replicate the experiments described here

In order to run these scripts you would need access to a few pickle files containining all the features. Email somandep@usc.edu to access these features.