Tianyang Zhao, Kunwar Yashraj Singh, Srikar Appalaraju, Peng Tang, Ying Nian Wu, Li Erran Li. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

We propose DocFormerv2, a multi-modal transformer for Visual Document Understanding (VDU). The VDU domain entails understanding documents (beyond mere OCR predictions) e.g., extracting information from a form, VQA for documents and other tasks. VDU is challenging as it needs a model to make sense of multiple modalities (visual, language and spatial) to make a prediction. Our approach, termed DocFormerv2 is an encoder-decoder transformer which takes as input - vision, language and spatial features. DocFormerv2 is pre-trained with unsupervised tasks employed asymmetrically i.e., two novel document tasks on encoder and one on the auto-regressive decoder. The unsupervised tasks have been carefully designed to ensure that the pre-training encourages local-feature alignment between multiple modalities. DocFormerv2 when evaluated on nine challenging datasets shows state-of-the-art performance on all over strong baselines - On TabFact (+4.3%), InfoVQA (+1.4%), FUNSD (+1.0%). Furthermore, to show generalization capabilities, on three VQA tasks involving scene-text, DocFormerv2 outperforms previous comparably-sized models and even does better than much larger models (such as GIT2, PaLI and Flamingo) on these tasks. Extensive ablations show that due to its novel pre-training tasks, DocFormerv2 understands multiple modalities better than prior-art in VDU.

Knowledge distillation aims at reducing model size without compromising much performance. Recent work has applied it to large vision-language (VL) Transformers, and has shown that attention maps in the multi-head attention modules of vision-language Transformers contain extensive intra-modal and cross-modal co-reference relations to be distilled. The standard approach is to apply a one-to-one attention map distillation loss, i.e. the Teacher's first attention head instructs the Student's first head, the second teaches the second, and so forth, but this only works when the numbers of attention heads in the Teacher and Student are the same. To remove this constraint, we propose a new Attention Map Alignment Distillation (AMAD) method for Transformers with multi-head attention, which works for a Teacher and a Student with different numbers of attention heads. Specifically, we soft-align different heads in Teacher and Student attention maps using a cosine similarity weighting. The Teacher head contributes more to the Student heads for which it has a higher similarity weight. Each Teacher head contributes to all the Student heads by minimizing the divergence between the attention activation distributions for the soft-aligned heads. No head is left behind. This distillation approach operates like cross-attention. We experiment on distilling VL-T5 and BLIP, and apply AMAD loss on their T5, BERT, and ViT sub-modules. We show, under vision-language setting, that AMAD outperforms conventional distillation methods on VQA-2.0, COCO captioning, and Multi30K translation datasets. We further show that even without VL pre-training, the distilled VL-T5 models outperform corresponding VL pre-trained VL-T5 models that are further fine-tuned by ground-truth signals, and that fine-tuning distillation can also compensate to some degree for the absence of VL pre-training for BLIP models.

We propose Strongly Supervised pre-training with ScreenShots (S4) - a novel pre-training paradigm for Vision-Language Models using data from large-scale web screenshot rendering. Using web screenshots unlocks a treasure trove of visual and textual cues that are not present in using image-text pairs. In S4, we leverage the inherent tree-structured hierarchy of HTML elements and the spatial localization to carefully design 10 pre-training tasks with large scale annotated data. These tasks resemble down- stream tasks across different domains and the annotations are cheap to obtain. We demonstrate that, compared to current screenshot pre-training objectives, our innovative pre-training method significantly enhances performance of image-to-text model in nine varied and popular downstream tasks - up to 76.1% improvements on Table Detection, and at least 1 % on Widget Captioning.

Sungnyun Kim, Haofu Liao, Srikar Appalaraju, Peng Tang, Zhuowen Tu, Ravi Kumar Satzoda, R. Manmatha, Vijay Mahadevan, Stefano Soatto. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

We propose a novel multimodal architecture for Scene Text Visual Question Answering (STVQA), named Layout-Aware Transformer (LaTr). The task of STVQA requires models to reason over different modalities. Thus, we first investigate the impact of each modality, and reveal the importance of the language module, especially when enriched with layout information. Accounting for this, we propose a single objective pre-training scheme that requires only text and spatial cues. We show that applying this pre-training scheme on scanned documents has certain advantages over using natural images, despite the domain gap. Scanned documents are easy to procure, text-dense and have a variety of layouts, helping the model learn various spatial cues (e.g. left-of, below etc.) by tying together language and layout information. Compared to existing approaches, our method performs vocabulary-free decoding and, as shown, generalizes well beyond the training vocabulary. We further demonstrate that LaTr improves robustness towards OCR errors, a common reason for failure cases in STVQA. In addition, by leveraging a vision transformer, we eliminate the need for an external object detector. LaTr outperforms state-of-the-art STVQA methods on multiple datasets. In particular, +7.6% on TextVQA, +10.8% on ST-VQA and +4.0% on OCR-VQA (all absolute accuracy numbers).

Recent advances in deep learning and computer vision have set new state of the art in logo recognition [2], [9], [36]. Logo recognition has mostly been approached as a closed-set object recognition problem and more recently as an open-set retrieval problem. Current approaches suffer from distinguishing visually similar logos, especially in open-set retrieval for very large-scale applications with thousands of brands. To address the problem, we propose a multi-task learning architecture of deep metric learning and scene text recognition. We use brand names as weak labels and enforce the model to simultaneously extract distinct visual features as well as predict brand name text. To achieve it, we collected a dataset with 3 Million logos cropped from Amazon Product Catalog images across nearly 8K brands, named PL8K. Our experiments show that adding the task of text recognition during training boosts the model’s retrieval performance both on our PL8K dataset and on five other public logo datasets.

We present DocFormer - a multi-modal transformer based architecture for the task of Visual Document Understanding (VDU). VDU is a challenging problem which aims to understand documents in their varied formats (forms, receipts etc.) and layouts. In addition, DocFormer is pre-trained in an unsupervised fashion using carefully designed tasks which encourage multi-modal interaction. DocFormer uses text, vision and spatial features and combines them using a novel multi-modal self-attention layer. DocFormer also shares learned spatial embeddings across modalities which makes it easy for the model to correlate text to visual tokens and vice versa. DocFormer is evaluated on 4 different datasets each with strong baselines. DocFormer achieves state-of-the-art results on all of them, sometimes beating models 4x its size (in no. of parameters).

This paper proposes a new end-to-end trainable model for lossy image compression, which includes several novel components. The method incorporates 1) an adequate perceptual similarity metric; 2) saliency in the images; 3) a hierarchical auto-regressive model. This paper demonstrates that the popularly used evaluations metrics such as MS-SSIM and PSNR are inadequate for judging the performance of image compression techniques as they do not align with the human perception of similarity. Alternatively, a new metric is proposed, which is learned on perceptual similarity data specific to image compression. The proposed compression model incorporates the salient regions and optimizes on the proposed perceptual similarity metric. The model not only generates images which are visually better but also gives superior performance for subsequent computer vision tasks such as object detection and segmentation when compared to existing engineered or learned compression techniques.

Logo recognition is the task of identifying and classifying logos. Logo recognition is a challenging problem as there is no clear definition of a logo and there are huge variations of logos, brands and re-training to cover every variation is impractical. In this paper, we formulate logo recognition as a few-shot object detection problem. The two main components in our pipeline are universal logo detector and few-shot logo recognizer. The universal logo detector is a class-agnostic deep object detector network which tries to learn the characteristics of what makes a logo. It predicts bounding boxes on likely logo regions. These logo regions are then classified by logo recognizer using nearest neighbor search, trained by triplet loss using proxies. We also annotated a first of its kind product logo dataset containing 2000 logos from 295K images collected from Amazon called PL2K. Our pipeline achieves 97% recall with 0.6 mAP on PL2K test dataset and state-of-the-art 0.565 mAP on the publicly available FlickrLogos-32 test set without fine-tuning.