Mutual Information Divergence: A Unified Metric for Multimodal Generative Models
This provides a more reliable evaluation metric for researchers and practitioners working on multimodal generative models, though it is incremental as it builds on existing CLIP-based approaches.
The paper tackles the problem of evaluating multimodal generative models like text-to-image generation and image captioning, which is complicated by continuous outputs and sampling techniques, by proposing Mutual Information Divergence (MID) as a unified metric using CLIP features. The result shows that MID significantly outperforms competing metrics in consistency, sample parsimony, and robustness across benchmarks.
Text-to-image generation and image captioning are recently emerged as a new experimental paradigm to assess machine intelligence. They predict continuous quantity accompanied by their sampling techniques in the generation, making evaluation complicated and intractable to get marginal distributions. Based on a recent trend that multimodal generative evaluations exploit a vison-and-language pre-trained model, we propose the negative Gaussian cross-mutual information using the CLIP features as a unified metric, coined by Mutual Information Divergence (MID). To validate, we extensively compare it with competing metrics using carefully-generated or human-annotated judgments in text-to-image generation and image captioning tasks. The proposed MID significantly outperforms the competitive methods by having consistency across benchmarks, sample parsimony, and robustness toward the exploited CLIP model. We look forward to seeing the underrepresented implications of the Gaussian cross-mutual information in multimodal representation learning and the future works based on this novel proposition.