Polos: Multimodal Metric Learning from Human Feedback for Image Captioning

Establishing an automatic evaluation metric that closely aligns with human judgments is essential for effectively developing image captioning models. Recent data-driven metrics have demonstrated a stronger correlation with human judgments than classic metrics such as CIDEr; however they lack sufficient capabilities to handle hallucinations and generalize across diverse images and texts partially because they compute scalar similarities merely using embeddings learned from tasks unrelated to image captioning evaluation. In this study, we propose Polos, a supervised automatic evaluation metric for image captioning models. Polos computes scores from multimodal inputs, using a parallel feature extraction mechanism that leverages embeddings trained through large-scale contrastive learning. To train Polos, we introduce Multimodal Metric Learning from Human Feedback (M$^2$LHF), a framework for developing metrics based on human feedback. We constructed the Polaris dataset, which comprises 131K human judgments from 550 evaluators, which is approximately ten times larger than standard datasets. Our approach achieved state-of-the-art performance on Composite, Flickr8K-Expert, Flickr8K-CF, PASCAL-50S, FOIL, and the Polaris dataset, thereby demonstrating its effectiveness and robustness.

Learning-To-Rank Approach for Identifying Everyday Objects Using a Physical-World Search Engine

Domestic service robots offer a solution to the increasing demand for daily care and support. A human-in-the-loop approach that combines automation and operator intervention is considered to be a realistic approach to their use in society. Therefore, we focus on the task of retrieving target objects from open-vocabulary user instructions in a human-in-the-loop setting, which we define as the learning-to-rank physical objects (LTRPO) task. For example, given the instruction "Please go to the dining room which has a round table. Pick up the bottle on it," the model is required to output a ranked list of target objects that the operator/user can select. In this paper, we propose MultiRankIt, which is a novel approach for the LTRPO task. MultiRankIt introduces the Crossmodal Noun Phrase Encoder to model the relationship between phrases that contain referring expressions and the target bounding box, and the Crossmodal Region Feature Encoder to model the relationship between the target object and multiple images of its surrounding contextual environment. Additionally, we built a new dataset for the LTRPO task that consists of instructions with complex referring expressions accompanied by real indoor environmental images that feature various target objects. We validated our model on the dataset and it outperformed the baseline method in terms of the mean reciprocal rank and recall@k. Furthermore, we conducted physical experiments in a setting where a domestic service robot retrieved everyday objects in a standardized domestic environment, based on users' instruction in a human--in--the--loop setting. The experimental results demonstrate that the success rate for object retrieval achieved 80%. Our code is available at https://github.com/keio-smilab23/MultiRankIt.

DialMAT: Dialogue-Enabled Transformer with Moment-Based Adversarial Training

This paper focuses on the DialFRED task, which is the task of embodied instruction following in a setting where an agent can actively ask questions about the task. To address this task, we propose DialMAT. DialMAT introduces Moment-based Adversarial Training, which incorporates adversarial perturbations into the latent space of language, image, and action. Additionally, it introduces a crossmodal parallel feature extraction mechanism that applies foundation models to both language and image. We evaluated our model using a dataset constructed from the DialFRED dataset and demonstrated superior performance compared to the baseline method in terms of success rate and path weighted success rate. The model secured the top position in the DialFRED Challenge, which took place at the CVPR 2023 Embodied AI workshop.

JaSPICE: Automatic Evaluation Metric Using Predicate-Argument Structures for Image Captioning Models

Image captioning studies heavily rely on automatic evaluation metrics such as BLEU and METEOR. However, such n-gram-based metrics have been shown to correlate poorly with human evaluation, leading to the proposal of alternative metrics such as SPICE for English; however, no equivalent metrics have been established for other languages. Therefore, in this study, we propose an automatic evaluation metric called JaSPICE, which evaluates Japanese captions based on scene graphs. The proposed method generates a scene graph from dependencies and the predicate-argument structure, and extends the graph using synonyms. We conducted experiments employing 10 image captioning models trained on STAIR Captions and PFN-PIC and constructed the Shichimi dataset, which contains 103,170 human evaluations. The results showed that our metric outperformed the baseline metrics for the correlation coefficient with the human evaluation.