Abstract: Self-Supervised learning from multimodal image and text data allows deep neural networks to learn powerful features with no need of human annotated data. Web and Social Media platforms provide a virtually unlimited amount of this multimodal data. In this work we propose to exploit this free available data to learn a multimodal image and text embedding, aiming to leverage the semantic knowledge learnt in the text domain and transfer it to a visual model for semantic image retrieval. We demonstrate that the proposed pipeline can learn from images with associated text without supervision and analyze the semantic structure of the learnt joint image and text embeddingspace. Weperformathoroughanalysisandperformancecomparisonoffivedifferentstateof the art text embeddings in three different benchmarks. We show that the embeddings learnt with Web and Social Media data have competitive performances over supervised methods in the text basedimageretrievaltask,andweclearlyoutperformstateoftheartintheMIRFlickrdatasetwhen training in the target data. Further, we demonstrate how semantic multimodal image retrieval can be performed using the learnt embeddings, going beyond classical instance-level retrieval problems. Finally, we present a new dataset, InstaCities1M, composed by Instagram images and their associated texts that can be used for fair comparison of image-text embeddings.

Abstract: End-to-end training from scratch of current deep architectures for new computer vision problems would require Imagenet-scale datasets, and this is not always possible. In this paper we present a method that is able to take advantage of freely available multi-modal content to train computer vision algorithms without human supervision. We put forward the idea of performing self-supervised learning of visual features by mining a large scale corpus of multi-modal (text and image) documents. We show that discriminative visual features can be learnt efficiently by training a CNN to predict the semantic context in which a particular image is more probable to appear as an illustration. For this we leverage the hidden semantic structures discovered in the text corpus with a well-known topic modeling technique. Our experiments demonstrate state of the art performance in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or natural-supervised approaches.

Abstract: Zero-shot sketch-based image retrieval (SBIR) is an emerging task in computer vision, allowing to retrieve natural images relevant to sketch queries that might not been seen in the training phase. Existing works either require aligned sketch-image pairs or inefficient memory fusion layer for mapping the visual information to a semantic space. In this work, we propose a semantically aligned paired cycle-consistent generative (SEM-PCYC) model for zero-shot SBIR, where each branch maps the visual information to a common semantic space via an adversarial training. Each of these branches maintains a cycle consistency that only requires supervision at category levels, and avoids the need of highly-priced aligned sketch-image pairs. A classification criteria on the generators' outputs ensures the visual to semantic space mapping to be discriminating. Furthermore, we propose to combine textual and hierarchical side information via a feature selection auto-encoder that selects discriminating side information within a same end-to-end model. Our results demonstrate a significant boost in zero-shot SBIR performance over the state-of-the-art on the challenging Sketchy and TU-Berlin datasets.

Abstract: Document Layout Analysis (DLA) is the process of automatically identifying and categorizing the structural components (e.g. Text, Figure, Table, etc.) within a document to extract meaningful content and establish the page's layout structure. It is a crucial stage in document parsing, contributing to their comprehension. However, traditional DLA approaches often demand a significant volume of labeled training data, and the labor-intensive task of generating high-quality annotated training data poses a substantial challenge. In order to address this challenge, we proposed a semi-supervised setting that aims to perform learning on limited annotated categories by eliminating exhaustive and expensive mask annotations. The proposed setting is expected to be generalizable to novel categories as it learns the underlying positional information through a support set and class information through Co-Occurrence that can be generalized from annotated categories to novel categories. Here, we first extract features from the input image and support set with a shared multi-scale feature acquisition backbone. Then, the extracted feature representation is fed to the transformer encoder as a query. Later on, we utilize a semantic embedding network before the decoder to capture the underlying semantic relationships and similarities between different instances, enabling the model to make accurate predictions or classifications with only a limited amount of labeled data. Extensive experimentation on competitive benchmarks like PRIMA, DocLayNet, and Historical Japanese (HJ) demonstrate that this generalized setup obtains significant performance compared to the conventional supervised approach.