Vincent Poulain D'Andecy

Rubèn Tito, Khanh Nguyen, Marlon Tobaben et al.

Document Visual Question Answering (DocVQA) has quickly grown into a central task of document understanding. But despite the fact that documents contain sensitive or copyrighted information, none of the current DocVQA methods offers strong privacy guarantees. In this work, we explore privacy in the domain of DocVQA for the first time, highlighting privacy issues in state of the art multi-modal LLM models used for DocVQA, and explore possible solutions. Specifically, we focus on invoice processing as a realistic document understanding scenario, and propose a large scale DocVQA dataset comprising invoice documents and associated questions and answers. We employ a federated learning scheme, that reflects the real-life distribution of documents in different businesses, and we explore the use case where the data of the invoice provider is the sensitive information to be protected. We demonstrate that non-private models tend to memorise, a behaviour that can lead to exposing private information. We then evaluate baseline training schemes employing federated learning and differential privacy in this multi-modal scenario, where the sensitive information might be exposed through either or both of the two input modalities: vision (document image) or language (OCR tokens). Finally, we design attacks exploiting the memorisation effect of the model, and demonstrate their effectiveness in probing a representative DocVQA models.

Eliott Thomas, Mickael Coustaty, Aurelie Joseph et al.

Extracting tables from documents is a critical task across various industries, especially on business documents like invoices and reports. Existing systems based on DEtection TRansformer (DETR) such as TAble TRansformer (TATR), offer solutions for Table Detection (TD) and Table Structure Recognition (TSR) but face challenges with diverse table formats and common errors like incorrect area detection and overlapping columns. This research introduces RAPTOR, a modular post-processing system designed to enhance state-of-the-art models for improved table extraction, particularly for product tables. RAPTOR addresses recurrent TD and TSR issues, improving both precision and structural predictions. For TD, we use DETR (trained on ICDAR 2019) and TATR (trained on PubTables-1M and FinTabNet), while TSR only relies on TATR. A Genetic Algorithm is incorporated to optimize RAPTOR's module parameters, using a private dataset of product tables to align with industrial needs. We evaluate our method on two private datasets of product tables, the public DOCILE dataset (which contains tables similar to our target product tables), and the ICDAR 2013 and ICDAR 2019 datasets. The results demonstrate that while our approach excels at product tables, it also maintains reasonable performance across diverse table formats. An ablation study further validates the contribution of each module in our system.

Marlon Tobaben, Mohamed Ali Souibgui, Rubèn Tito et al.

The Privacy Preserving Federated Learning Document VQA (PFL-DocVQA) competition challenged the community to develop provably private and communication-efficient solutions in a federated setting for a real-life use case: invoice processing. The competition introduced a dataset of real invoice documents, along with associated questions and answers requiring information extraction and reasoning over the document images. Thereby, it brings together researchers and expertise from the document analysis, privacy, and federated learning communities. Participants fine-tuned a pre-trained, state-of-the-art Document Visual Question Answering model provided by the organizers for this new domain, mimicking a typical federated invoice processing setup. The base model is a multi-modal generative language model, and sensitive information could be exposed through either the visual or textual input modality. Participants proposed elegant solutions to reduce communication costs while maintaining a minimum utility threshold in track 1 and to protect all information from each document provider using differential privacy in track 2. The competition served as a new testbed for developing and testing private federated learning methods, simultaneously raising awareness about privacy within the document image analysis and recognition community. Ultimately, the competition analysis provides best practices and recommendations for successfully running privacy-focused federated learning challenges in the future.

Eliott Thomas, Mickael Coustaty, Aurelie Joseph et al.

Automating table extraction (TE) from business documents is critical for industrial workflows but remains challenging due to sparse annotations and error-prone multi-stage pipelines. While semi-supervised learning (SSL) can leverage unlabeled data, existing methods rely on confidence scores that poorly reflect extraction quality. We propose QUEST, a Quality-aware Semi-supervised Table extraction framework designed for business documents. QUEST introduces a novel quality assessment model that evaluates structural and contextual features of extracted tables, trained to predict F1 scores instead of relying on confidence metrics. This quality-aware approach guides pseudo-label selection during iterative SSL training, while diversity measures (DPP, Vendi score, IntDiv) mitigate confirmation bias. Experiments on a proprietary business dataset (1000 annotated + 10000 unannotated documents) show QUEST improves F1 from 64% to 74% and reduces empty predictions by 45% (from 12% to 6.5%). On the DocILE benchmark (600 annotated + 20000 unannotated documents), QUEST achieves a 50% F1 score (up from 42%) and reduces empty predictions by 19% (from 27% to 22%). The framework's interpretable quality assessments and robustness to annotation scarcity make it particularly suited for business documents, where structural consistency and data completeness are paramount.

Abdel Belaïd, K. C. Santosh, Vincent Poulain D'Andecy

The aim of the paper is to separate handwritten and printed text from a real document embedded with noise, graphics including annotations. Relying on run-length smoothing algorithm (RLSA), the extracted pseudo-lines and pseudo-words are used as basic blocks for classification. To handle this, a multi-class support vector machine (SVM) with Gaussian kernel performs a first labelling of each pseudo-word including the study of local neighbourhood. It then propagates the context between neighbours so that we can correct possible labelling errors. Considering running time complexity issue, we propose linear complexity methods where we use k-NN with constraint. When using a kd-tree, it is almost linearly proportional to the number of pseudo-words. The performance of our system is close to 90%, even when very small learning dataset where samples are basically composed of complex administrative documents.