End-to-end Image Captioning Exploits Multimodal Distributional Similarity
arXiv:1809.04144
Abstract
We hypothesize that end-to-end neural image captioning systems work seemingly well because they exploit and learn `distributional similarity' in a multimodal feature space by mapping a test image to similar training images in this space and generating a caption from the same space. To validate our hypothesis, we focus on the `image' side of image captioning, and vary the input image representation but keep the RNN text generation component of a CNN-RNN model constant. Our analysis indicates that image captioning models (i) are capable of separating structure from noisy input representations; (ii) suffer virtually no significant performance loss when a high dimensional representation is compressed to a lower dimensional space; (iii) cluster images with similar visual and linguistic information together. Our findings indicate that our distributional similarity hypothesis holds. We conclude that regardless of the image representation used image captioning systems seem to match images and generate captions in a learned joint image-text semantic subspace.
arXiv commentPublished in BMVC 2018
References in corpus (10)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Sequence to Sequence Learning with Neural Networks
- DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition
- Recurrent Neural Network Regularization
- Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)
- Microsoft COCO Captions: Data Collection and Evaluation Server
- CNN Features off-the-shelf: an Astounding Baseline for Recognition
- Are You Talking to a Machine? Dataset and Methods for Multilingual Image Question Answering
- Language Models for Image Captioning: The Quirks and What Works
- OBJ2TEXT: Generating Visually Descriptive Language from Object Layouts