Teakgyu Hong

Donghyun Kim, Teakgyu Hong, Moonbin Yim et al.

In recent years, research on visual document understanding (VDU) has grown significantly, with a particular emphasis on the development of self-supervised learning methods. However, one of the significant challenges faced in this field is the limited availability of publicly accessible visual corpora or extensive collections of images with detailed text annotations, particularly for non-Latin or resource-scarce languages. To address this challenge, we propose Web-based Visual Corpus Builder (Webvicob), a dataset generator engine capable of constructing large-scale, multilingual visual corpora from raw Wikipedia HTML dumps. Our experiments demonstrate that the data generated by Webvicob can be used to train robust VDU models that perform well on various downstream tasks, such as DocVQA and post-OCR parsing. Furthermore, when using a dataset of 1 million images generated by Webvicob, we observed an improvement of over 13% on the DocVQA Task 3 compared to a dataset of 11 million images from the IIT-CDIP. The implementation of our engine is publicly available on https://github.com/clovaai/webvicob

Minbyul Jeong, Jungho Cho, Minsoo Khang et al.

System messages play a crucial role in interactions with large language models (LLMs), often serving as prompts to initiate conversations. Through system messages, users can assign specific roles, perform intended tasks, incorporate background information, and specify various output formats and communication styles. Despite such versatility, publicly available datasets often lack system messages and are subject to strict license constraints in industrial applications. Moreover, manually annotating system messages that align with user instructions is resource-intensive. In light of these challenges, we introduce SysGen, a pipeline for generating system messages that better align assistant responses with user instructions using existing supervised fine-tuning datasets that lack system messages. Training open-source models on SysGen data yields substantial improvements in both single-turn (Multifacet) and multi-turn (SysBench) conversation benchmarks. Notably, our method shows strong gains in shorter conversations, suggesting that it enhances early-stage interaction effectiveness. Our qualitative analysis further emphasizes the value of diverse and structured system messages in improving LLM adaptability across varied user scenarios.

Geewook Kim, Teakgyu Hong, Moonbin Yim et al.

Understanding document images (e.g., invoices) is a core but challenging task since it requires complex functions such as reading text and a holistic understanding of the document. Current Visual Document Understanding (VDU) methods outsource the task of reading text to off-the-shelf Optical Character Recognition (OCR) engines and focus on the understanding task with the OCR outputs. Although such OCR-based approaches have shown promising performance, they suffer from 1) high computational costs for using OCR; 2) inflexibility of OCR models on languages or types of document; 3) OCR error propagation to the subsequent process. To address these issues, in this paper, we introduce a novel OCR-free VDU model named Donut, which stands for Document understanding transformer. As the first step in OCR-free VDU research, we propose a simple architecture (i.e., Transformer) with a pre-training objective (i.e., cross-entropy loss). Donut is conceptually simple yet effective. Through extensive experiments and analyses, we show a simple OCR-free VDU model, Donut, achieves state-of-the-art performances on various VDU tasks in terms of both speed and accuracy. In addition, we offer a synthetic data generator that helps the model pre-training to be flexible in various languages and domains. The code, trained model and synthetic data are available at https://github.com/clovaai/donut.

Teakgyu Hong, Donghyun Kim, Mingi Ji et al.

Key information extraction (KIE) from document images requires understanding the contextual and spatial semantics of texts in two-dimensional (2D) space. Many recent studies try to solve the task by developing pre-trained language models focusing on combining visual features from document images with texts and their layout. On the other hand, this paper tackles the problem by going back to the basic: effective combination of text and layout. Specifically, we propose a pre-trained language model, named BROS (BERT Relying On Spatiality), that encodes relative positions of texts in 2D space and learns from unlabeled documents with area-masking strategy. With this optimized training scheme for understanding texts in 2D space, BROS shows comparable or better performance compared to previous methods on four KIE benchmarks (FUNSD, SROIE*, CORD, and SciTSR) without relying on visual features. This paper also reveals two real-world challenges in KIE tasks-(1) minimizing the error from incorrect text ordering and (2) efficient learning from fewer downstream examples-and demonstrates the superiority of BROS over previous methods. Code is available at https://github.com/clovaai/bros.

Minsoo Khang, Teakgyu Hong

Table Structure Recognition (TSR) is a task aimed at converting table images into a machine-readable format (e.g. HTML), to facilitate other applications such as information retrieval. Recent works tackle this problem by identifying the HTML tags and text regions, where the latter is used for text extraction from the table document. These works however, suffer from misalignment issues when mapping text into the identified text regions. In this paper, we introduce a new TSR framework, called TFLOP (TSR Framework with LayOut Pointer mechanism), which reformulates the conventional text region prediction and matching into a direct text region pointing problem. Specifically, TFLOP utilizes text region information to identify both the table's structure tags and its aligned text regions, simultaneously. Without the need for region prediction and alignment, TFLOP circumvents the additional text region matching stage, which requires finely-calibrated post-processing. TFLOP also employs span-aware contrastive supervision to enhance the pointing mechanism in tables with complex structure. As a result, TFLOP achieves the state-of-the-art performance across multiple benchmarks such as PubTabNet, FinTabNet, and SynthTabNet. In our extensive experiments, TFLOP not only exhibits competitive performance but also shows promising results on industrial document TSR scenarios such as documents with watermarks or in non-English domain.

Yunsung Lee, Teakgyu Hong, Han-Cheol Cho et al.

Recently, self-supervised methods show remarkable achievements in image-level representation learning. Nevertheless, their image-level self-supervisions lead the learned representation to sub-optimal for dense prediction tasks, such as object detection, instance segmentation, etc. To tackle this issue, several recent self-supervised learning methods have extended image-level single embedding to pixel-level dense embeddings. Unlike image-level representation learning, due to the spatial deformation of augmentation, it is difficult to sample pixel-level positive pairs. Previous studies have sampled pixel-level positive pairs using the winner-takes-all among similarity or thresholding warped distance between dense embeddings. However, these naive methods can be struggled by background clutter and outliers problems. In this paper, we introduce Hough Contrastive Learning (HoughCL), a Hough space based method that enforces geometric consistency between two dense features. HoughCL achieves robustness against background clutter and outliers. Furthermore, compared to baseline, our dense positive pairing method has no additional learnable parameters and has a small extra computation cost. Compared to previous works, our method shows better or comparable performance on dense prediction fine-tuning tasks.