Fast Cheap Image Captioning Model Trained on Video Frames