Self-Annotated Training for Controllable Image Captioning
This addresses a training bottleneck for controllable image captioning, enabling better alignment with structural constraints like sentence length and part-of-speech sequences, which is incremental but specific to this domain.
The paper tackled the problem of training controllable image captioning models for structure-related control signals, where existing reinforcement methods fail due to focusing on content rather than structure, and proposed Self-Annotated Training (SAT) with a recursive annotation mechanism and alignment reward, resulting in improved accuracy and controllability as demonstrated on the MSCOCO benchmark.
The Controllable Image Captioning (CIC) task aims to generate captions conditioned on designated control signals. Several structure-related control signals are proposed to control the semantic structure of sentences, such as sentence length and Part-of-Speech tag sequences. However, due to the fact that the accuracy-based reward focuses mainly on contents rather than semantic structures, existing reinforcement training methods are not applicable to structure-related CIC models. The lack of reinforcement training leads to exposure bias and the inconsistency between the optimizing function and evaluation metrics. In this paper, we propose a novel reinforcement training method for structure-related control signals: Self-Annotated Training (SAT), to improve both the accuracy and controllability of CIC models. In SAT, a recursive annotation mechanism (RAM) is designed to force the input control signal to match the actual output sentence. Moreover, we propose an extra alignment reward to finetune the CIC model trained after SAT method, which further enhances the controllability of models. On the MSCOCO benchmark, we conduct extensive experiments on different structure-related control signals and on different baseline models, the results of which demonstrate the effectiveness and generalizability of our methods.