These methods typically involve largescale Scene-Text Visual Question Answering (ST-VQA) aims to pretraining followed by fine-tuning to adapt the model for understand scene text in images and answer questions related question-answering tasks in text-rich scene images, often ignoring to the text content. Most existing methods heavily rely on the the inevitable OCR text recognition challenges. In practice, accuracy of Optical Character Recognition (OCR) systems, scene images may exhibit phenomena such as blurring, and aggressive fine-tuning based on limited spatial location distortion, skewness, or uneven lighting, leading to erroneous information and erroneous OCR text information often leads character recognition by OCR systems, especially in cases to inevitable overfitting. In this paper, we propose a multimodal of low-quality handwriting. Even when OCR systems correctly adversarial training architecture with spatial awareness identify characters, discrete and semantically irrelevant capabilities. Specifically, we introduce an Adversarial OCR recognition results may impact the comprehension of the OCR Enhancement (AOE) module, which leverages adversarial text semantics.

Abstract:In visual search, the gallery set could be incrementally growing and added to the database in practice. However, existing methods rely on the model trained on the entire dataset, ignoring the continual updating of the model. Besides, as the model updates, the new model must re-extract features for the entire gallery set to maintain compatible feature space, imposing a high computational cost for a large gallery set. To address the issues of long-term visual search, we introduce a continual learning (CL) approach that can handle the incrementally growing gallery set with backward embedding consistency. We enforce the losses of inter-session data coherence, neighbor-session model coherence, and intra-session discrimination to conduct a continual learner.