Gaurav Harit

Somraj Gautam, Anathapindika Dravichi, Gaurav Harit

We introduce INDOTABVQA, a benchmark for evaluating cross-lingual Table Visual Question Answering (VQA) on real-world document images in Bahasa Indonesia. The dataset comprises 1,593 document images across three visual styles (bordered, borderless, and colorful) with one or more than one tables, and 1,593 question-answer sets in four languages: Bahasa Indonesia, English, Hindi, and Arabic. This enables evaluation of Vision-Language Models (VLMs) in both monolingual (Bahasa documents with Bahasa questions) and cross-lingual settings (Bahasa documents with questions in other languages). We benchmark leading open-source VLMs (Qwen2.5-VL, Gemma-3, LLaMA-3.2) and GPT-4o and reveal substantial performance gaps, particularly on structurally complex tables and in low-resource languages. Fine-tuning a compact 3B and LoRA-finetuned 7B model on our dataset yields 11.6% and 17.8% improvements in accuracy. Providing explicit table region coordinates as additional input further improves performance by 4-7%, demonstrating the value of Spatial priors for table-based reasoning. Our findings underscore the importance of language-diverse, domain-specific datasets and demonstrate that targeted fine-tuning can significantly enhance VLM performance on specialized document understanding tasks. INDOTABVQA provides a valuable resource for advancing research in cross-lingual, structure-aware document understanding, especially in underrepresented regions of the world. Full dataset can be accessed in huggingface at: https://huggingface.co/datasets/NusaBharat/INDOTABVQA}

Sagar Chakraborty, Gaurav Harit, Saptarshi Ghosh

State-of-the-art offline Optical Character Recognition (OCR) frameworks perform poorly on semi-structured handwritten domain-specific documents due to their inability to localize and label form fields with domain-specific semantics. Existing techniques for semi-structured document analysis have primarily used datasets comprising invoices, purchase orders, receipts, and identity-card documents for benchmarking. In this work, we build the first semi-structured document analysis dataset in the legal domain by collecting a large number of First Information Report (FIR) documents from several police stations in India. This dataset, which we call the FIR dataset, is more challenging than most existing document analysis datasets, since it combines a wide variety of handwritten text with printed text. We also propose an end-to-end framework for offline processing of handwritten semi-structured documents, and benchmark it on our novel FIR dataset. Our framework used Encoder-Decoder architecture for localizing and labelling the form fields and for recognizing the handwritten content. The encoder consists of Faster-RCNN and Vision Transformers. Further the Transformer-based decoder architecture is trained with a domain-specific tokenizer. We also propose a post-correction method to handle recognition errors pertaining to the domain-specific terms. Our proposed framework achieves state-of-the-art results on the FIR dataset outperforming several existing models

Somraj Gautam, Abhirama Subramanyam Penamakuri, Abhishek Bhandari et al.

We introduce MMCRICBENCH-3K, a benchmark for Visual Question Answering (VQA) on cricket scorecards, designed to evaluate large vision-language models (LVLMs) on complex numerical and cross-lingual reasoning over semi-structured tabular images. MMCRICBENCH-3K comprises 1,463 synthetically generated scorecard images from ODI, T20, and Test formats, accompanied by 1,500 English QA pairs. It includes two subsets: MMCRICBENCH-E-1.5K, featuring English scorecards, and MMCRICBENCH-H-1.5K, containing visually similar Hindi scorecards, with all questions and answers kept in English to enable controlled cross-script evaluation. The task demands reasoning over structured numerical data, multi-image context, and implicit domain knowledge. Empirical results show that even state-of-the-art LVLMs, such as GPT-4o and Qwen2.5VL, struggle on the English subset despite it being their primary training language and exhibit a further drop in performance on the Hindi subset. This reveals key limitations in structure-aware visual text understanding, numerical reasoning, and cross-lingual generalization. The dataset is publicly available via Hugging Face at https://huggingface.co/datasets/DIALab/MMCricBench, to promote LVLM research in this direction.

Somraj Gautam, Nachiketa Purohit, Gaurav Harit

Efficient data annotation remains a critical challenge in machine learning, particularly for object detection tasks requiring extensive labeled data. Active learning (AL) has emerged as a promising solution to minimize annotation costs by selecting the most informative samples. While traditional AL approaches primarily rely on uncertainty-based selection, recent advances suggest that incorporating diversity-based strategies can enhance sampling efficiency in object detection tasks. Our approach ensures the selection of representative examples that improve model generalization. We evaluate our method on two benchmark datasets (TableBank-LaTeX, TableBank-Word) using state-of-the-art table detection architectures, CascadeTabNet and YOLOv9. Our results demonstrate that AL-based example selection significantly outperforms random sampling, reducing annotation effort given a limited budget while maintaining comparable performance to fully supervised models. Our method achieves higher mAP scores within the same annotation budget.

Arka Ujjal Dey, Ernest Valveny, Gaurav Harit

The open-ended question answering task of Text-VQA often requires reading and reasoning about rarely seen or completely unseen scene-text content of an image. We address this zero-shot nature of the problem by proposing the generalized use of external knowledge to augment our understanding of the scene text. We design a framework to extract, validate, and reason with knowledge using a standard multimodal transformer for vision language understanding tasks. Through empirical evidence and qualitative results, we demonstrate how external knowledge can highlight instance-only cues and thus help deal with training data bias, improve answer entity type correctness, and detect multiword named entities. We generate results comparable to the state-of-the-art on three publicly available datasets, under the constraints of similar upstream OCR systems and training data.

Hiteshi Jain, Gaurav Harit, Avinash Sharma

Automated vision-based score estimation models can be used as an alternate opinion to avoid judgment bias. In the past works the score estimation models were learned by regressing the video representations to the ground truth score provided by the judges. However such regression-based solutions lack interpretability in terms of giving reasons for the awarded score. One solution to make the scores more explicable is to compare the given action video with a reference video. This would capture the temporal variations w.r.t. the reference video and map those variations to the final score. In this work, we propose a new action scoring system as a two-phase system: (1) A Deep Metric Learning Module that learns similarity between any two action videos based on their ground truth scores given by the judges; (2) A Score Estimation Module that uses the first module to find the resemblance of a video to a reference video in order to give the assessment score. The proposed scoring model has been tested for Olympics Diving and Gymnastic vaults and the model outperforms the existing state-of-the-art scoring models.

Arka Ujjal Dey, Suman Kumar Ghosh, Ernest Valveny et al.

Images with visual and scene text content are ubiquitous in everyday life. However, current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper, we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We do not only extract and encode visual and scene text cues, but also model their interplay to generate a contextual joint embedding with richer semantics. The contextual embedding thus generated is applied to retrieval and classification tasks on multimedia images, with scene text content, to demonstrate its effectiveness. In the retrieval framework, we augment our learned text-visual semantic representation with scene text cues, to mitigate vocabulary misses that may have occurred during the semantic embedding. To deal with irrelevant or erroneous recognition of scene text, we also apply query-based attention to our text channel. We show how the multi-channel approach, involving visual semantics and scene text, improves upon state of the art.

Anukriti Bansal, Sumantra Dutta Roy, Gaurav Harit

Layouts and sub-layouts constitute an important clue while searching a document on the basis of its structure, or when textual content is unknown/irrelevant. A sub-layout specifies the arrangement of document entities within a smaller portion of the document. We propose an efficient graph-based matching algorithm, integrated with hash-based indexing, to prune a possibly large search space. A user can specify a combination of sub-layouts of interest using sketch-based queries. The system supports partial matching for unspecified layout entities. We handle cases of segmentation pre-processing errors (for text/non-text blocks) with a symmetry maximization-based strategy, and accounting for multiple domain-specific plausible segmentation hypotheses. We show promising results of our system on a database of unstructured entities, containing 4776 newspaper images.

Pratik Kalshetti, Manas Bundele, Parag Rahangdale et al.

Image segmentation is often performed on medical images for identifying diseases in clinical evaluation. Hence it has become one of the major research areas. Conventional image segmentation techniques are unable to provide satisfactory segmentation results for medical images as they contain irregularities. They need to be pre-processed before segmentation. In order to obtain the most suitable method for medical image segmentation, we propose a two stage algorithm. The first stage automatically generates a binary marker image of the region of interest using mathematical morphology. This marker serves as the mask image for the second stage which uses GrabCut on the input image thus resulting in an efficient segmented result. The obtained result can be further refined by user interaction which can be done using the Graphical User Interface (GUI). Experimental results show that the proposed method is accurate and provides satisfactory segmentation results with minimum user interaction on medical as well as natural images.