The conventional training approach for image captioning involves pre-training a network using teacher forcing and subsequent fine-tuning with Self-Critical Sequence Training to maximize hand-crafted captioning metrics. However, when attempting to optimize modern and higher-quality metrics like CLIP-Score and PAC-Score, this training method often encounters instability and fails to acquire the genuine descriptive capabilities needed to produce fluent and informative captions. In this paper, we propose a new training paradigm termed Direct CLIP-Based Optimization (DiCO). Our approach jointly learns and optimizes a reward model that is distilled from a learnable captioning evaluator with high human correlation. This is done by solving a weighted classification problem directly inside the captioner. At the same time, DiCO prevents divergence from the original model, ensuring that fluency is maintained. DiCO not only exhibits improved stability and enhanced quality in the generated captions but also aligns more closely with human preferences compared to existing methods, especially in modern metrics. Additionally, it maintains competitive performance in traditional metrics. Our source code and trained models are publicly available at https://github.com/aimagelab/DiCO.

The goal of fine-grained image description generation techniques is to learn detailed information from images and simulate human-like descriptions that provide coherent and comprehensive textual details about the image content. Currently, most of these methods face two main challenges: description repetition and omission. Moreover, the existing evaluation metrics cannot clearly reflect the performance of models on these two issues. To address these challenges, we propose an innovative Fine-grained Image Description Generation model based on Joint Objectives. Furthermore, we introduce new object-based evaluation metrics to more intuitively assess the model's performance in handling description repetition and omission. This novel approach combines visual features at both the image level and object level to maximize their advantages and incorporates an object penalty mechanism to reduce description repetition. Experimental results demonstrate that our proposed method significantly improves the CIDEr evaluation metric, indicating its excellent performance in addressing description repetition and omission issues.

Deep neural networks have achieved promising results in automatic image captioning due to their effective representation learning and context-based content generation capabilities. As a prominent type of deep features used in many of the recent image captioning methods, the well-known bottomup features provide a detailed representation of different objects of the image in comparison with the feature maps directly extracted from the raw image. However, the lack of high-level semantic information about the relationships between these objects is an important drawback of bottom-up features, despite their expensive and resource-demanding extraction procedure. To take advantage of visual relationships in caption generation, this paper proposes a deep neural network architecture for image captioning based on fusing the visual relationships information extracted from an image's scene graph with the spatial feature maps of the image. A multi-modal reward function is then introduced for deep reinforcement learning of the proposed network using a combination of language and vision similarities in a common embedding space. The results of extensive experimentation on the MSCOCO dataset show the effectiveness of using visual relationships in the proposed captioning method. Moreover, the results clearly indicate that the proposed multi-modal reward in deep reinforcement learning leads to better model optimization, outperforming several state-of-the-art image captioning algorithms, while using light and easy to extract image features. A detailed experimental study of the components constituting the proposed method is also presented.

Large-scale vision-language pre-trained (VLP) models are prone to hallucinate non-existent visual objects when generating text based on visual information. In this paper, we systematically study the object hallucination problem from three aspects. First, we examine recent state-of-the-art VLP models, showing that they still hallucinate frequently, and models achieving better scores on standard metrics (e.g., CIDEr) could be more unfaithful. Second, we investigate how different types of image encoding in VLP influence hallucination, including region-based, grid-based, and patch-based. Surprisingly, we find that patch-based features perform the best and smaller patch resolution yields a non-trivial reduction in object hallucination. Third, we decouple various VLP objectives and demonstrate that token-level image-text alignment and controlled generation are crucial to reducing hallucination. Based on that, we propose a simple yet effective VLP loss named ObjMLM to further mitigate object hallucination. Results show that it reduces object hallucination by up to 17.4% when tested on two benchmarks (COCO Caption for in-domain and NoCaps for out-of-domain evaluation).

In this paper we propose a new conditional GAN for image captioning that enforces semantic alignment between images and captions through a co-attentive discriminator and a context-aware LSTM sequence generator. In order to train these sequence GANs, we empirically study two algorithms: Self-critical Sequence Training (SCST) and Gumbel Straight-Through. Both techniques are confirmed to be viable for training sequence GANs. However, SCST displays better gradient behavior despite not directly leveraging gradients from the discriminator. This ensures a stronger stability of sequence GANs training and ultimately produces models with improved results under human evaluation. Automatic evaluation of GAN trained captioning models is an open question. To remedy this, we introduce a new semantic score with strong correlation to human judgement. As a paradigm for evaluation, we suggest that the generalization ability of the captioner to Out of Context (OOC) scenes is an important criterion to assess generalization and composition. To this end, we propose an OOC dataset which, combined with our automatic metric of semantic score, is a new benchmark for the captioning community to measure the generalization ability of automatic image captioning. Under this new OOC benchmark, and on the traditional MSCOCO dataset, our models trained with SCST have strong performance in both semantic score and human evaluation.

Recently it has been shown that policy-gradient methods for reinforcement learning can be utilized to train deep end-to-end systems directly on non-differentiable metrics for the task at hand. In this paper we consider the problem of optimizing image captioning systems using reinforcement learning, and show that by carefully optimizing our systems using the test metrics of the MSCOCO task, significant gains in performance can be realized. Our systems are built using a new optimization approach that we call self-critical sequence training (SCST). SCST is a form of the popular REINFORCE algorithm that, rather than estimating a "baseline" to normalize the rewards and reduce variance, utilizes the output of its own test-time inference algorithm to normalize the rewards it experiences. Using this approach, estimating the reward signal (as actor-critic methods must do) and estimating normalization (as REINFORCE algorithms typically do) is avoided, while at the same time harmonizing the model with respect to its test-time inference procedure. Empirically we find that directly optimizing the CIDEr metric with SCST and greedy decoding at test-time is highly effective. Our results on the MSCOCO evaluation sever establish a new state-of-the-art on the task, improving the best result in terms of CIDEr from 104.9 to 114.7.