Publications by Lorenzo Baraldi

Deep learning-based approaches to Handwritten Text Recognition (HTR) have shown remarkable results on publicly available large datasets, both modern and historical. However, it is often the case that historical manuscripts are preserved in small collections, most of the time with unique characteristics in terms of paper support, author handwriting style, and language. State-of-the-art HTR approaches struggle to obtain good performance on such small manuscript collections, for which few training samples are available. In this paper, we focus on HTR on small historical datasets and propose a new historical dataset, which we call Leopardi, with the typical characteristics of small manuscript collections, consisting of letters by the poet Giacomo Leopardi, and devise strategies to deal with the training data scarcity scenario. In particular, we explore the use of carefully designed but cost-effective synthetic data for pre-training HTR models to be applied to small single-author manuscripts. Extensive experiments validate the suitability of the proposed approach, and both the Leopardi dataset and synthetic data will be available to favor further research in this direction.

Image captioning models have lately shown impressive results when applied to standard datasets. Switching to real-life scenarios, however, constitutes a challenge due to the larger variety of visual concepts which are not covered in existing training sets. For this reason, novel object captioning (NOC) has recently emerged as a paradigm to test captioning models on objects which are unseen during the training phase. In this paper, we present a novel approach for NOC that learns to select the most relevant objects of an image, regardless of their adherence to the training set, and to constrain the generative process of a language model accordingly. Our architecture is fully-attentive and end-to-end trainable, also when incorporating constraints. We perform experiments on the held-out COCO dataset, where we demonstrate improvements over the state of the art, both in terms of adaptability to novel objects and caption quality.

Vision-and-Language Navigation (VLN) is a challenging task in which an agent needs to follow a language-specified path to reach a target destination. The goal gets even harder as the actions available to the agent get simpler and move towards low-level, atomic interactions with the environment. This setting takes the name of low-level VLN. In this paper, we strive for the creation of an agent able to tackle three key issues: multi-modality, long-term dependencies, and adaptability towards different locomotive settings. To that end, we devise "Perceive, Transform, and Act" (PTA): a fully-attentive VLN architecture that leaves the recurrent approach behind and the first Transformer-like architecture incorporating three different modalities -- natural language, images, and low-level actions for the agent control. In particular, we adopt an early fusion strategy to merge lingual and visual information efficiently in our encoder. We then propose to refine the decoding phase with a late fusion extension between the agent's history of actions and the perceptual modalities. We experimentally validate our model on two datasets: PTA achieves promising results in low-level VLN on R2R and achieves good performance in the recently proposed R4R benchmark. Our code is publicly available at https://github.com/aimagelab/perceive-transform-and-act.

The research field of Embodied AI has witnessed substantial progress in visual navigation and exploration thanks to powerful simulating platforms and the availability of 3D data of indoor and photorealistic environments. These two factors have opened the doors to a new generation of intelligent agents capable of achieving nearly perfect PointGoal Navigation. However, such architectures are commonly trained with millions, if not billions, of frames and tested in simulation. Together with great enthusiasm, these results yield a question: how many researchers will effectively benefit from these advances? In this work, we detail how to transfer the knowledge acquired in simulation into the real world. To that end, we describe the architectural discrepancies that damage the Sim2Real adaptation ability of models trained on the Habitat simulator and propose a novel solution tailored towards the deployment in real-world scenarios. We then deploy our models on a LoCoBot, a Low-Cost Robot equipped with a single Intel RealSense camera. Different from previous work, our testing scene is unavailable to the agent in simulation. The environment is also inaccessible to the agent beforehand, so it cannot count on scene-specific semantic priors. In this way, we reproduce a setting in which a research group (potentially from other fields) needs to employ the agent visual navigation capabilities as-a-Service. Our experiments indicate that it is possible to achieve satisfying results when deploying the obtained model in the real world.

As the request for deep learning solutions increases, the need for explainability is even more fundamental. In this setting, particular attention has been given to visualization techniques, that try to attribute the right relevance to each input pixel with respect to the output of the network. In this paper, we focus on Class Activation Mapping (CAM) approaches, which provide an effective visualization by taking weighted averages of the activation maps. To enhance the evaluation and the reproducibility of such approaches, we propose a novel set of metrics to quantify explanation maps, which show better effectiveness and simplify comparisons between approaches. To evaluate the appropriateness of the proposal, we compare different CAM-based visualization methods on the entire ImageNet validation set, fostering proper comparisons and reproducibility.

Action Detection is a complex task that aims to detect and classify human actions in video clips. Typically, it has been addressed by processing fine-grained features extracted from a video classification backbone. Recently, thanks to the robustness of object and people detectors, a deeper focus has been added on relationship modelling. Following this line, we propose a graph-based framework to learn high-level interactions between people and objects, in both space and time. In our formulation, spatio-temporal relationships are learned through self-attention on a multi-layer graph structure which can connect entities from consecutive clips, thus considering long-range spatial and temporal dependencies. The proposed module is backbone independent by design and does not require end-to-end training. Extensive experiments are conducted on the AVA dataset, where our model demonstrates state-of-the-art results and consistent improvements over baselines built with different backbones. Code is publicly available at https://github.com/aimagelab/STAGE_action_detection.

Handwritten Text Recognition (HTR) in free-layout pages is a valuable yet challenging task which aims to automatically understand handwritten texts. State-of-the-art approaches in this field usually encode input images with Convolutional Neural Networks, whose kernels are typically defined on a fixed grid and focus on all input pixels independently. However, this is in contrast with the sparse nature of handwritten pages, in which only pixels representing the ink of the writing are useful for the recognition task. Furthermore, the standard convolution operator is not explicitly designed to take into account the great variability in shape, scale, and orientation of handwritten characters. To overcome these limitations, we investigate the use of deformable convolutions for handwriting recognition. This type of convolution deform the convolution kernel according to the content of the neighborhood, and can therefore be more adaptable to geometric variations and other deformations of the text. Experiments conducted on the IAM and RIMES datasets demonstrate that the use of deformable convolutions is a promising direction for the design of novel architectures for handwritten text recognition.

Recurrent Neural Networks with Long Short-Term Memory (LSTM) make use of gating mechanisms to mitigate exploding and vanishing gradients when learning long-term dependencies. For this reason, LSTMs and other gated RNNs are widely adopted, being the standard de facto for many sequence modeling tasks. Although the memory cell inside the LSTM contains essential information, it is not allowed to influence the gating mechanism directly. In this work, we improve the gate potential by including information coming from the internal cell state. The proposed modification, named Working Memory Connection, consists in adding a learnable nonlinear projection of the cell content into the network gates. This modification can fit into the classical LSTM gates without any assumption on the underlying task, being particularly effective when dealing with longer sequences. Previous research effort in this direction, which goes back to the early 2000s, could not bring a consistent improvement over vanilla LSTM. As part of this paper, we identify a key issue tied to previous connections that heavily limits their effectiveness, hence preventing a successful integration of the knowledge coming from the internal cell state. We show through extensive experimental evaluation that Working Memory Connections constantly improve the performance of LSTMs on a variety of tasks. Numerical results suggest that the cell state contains useful information that is worth including in the gate structure.

Text-image retrieval has been recently becoming a hot-spot research field, thanks to the development of deeply-learnable architectures which can retrieve visual items given textual queries and vice-versa. The key idea of many state-of-the-art approaches has been that of learning a joint multi-modal embedding space in which text and images could be projected and compared. Here we take a different approach and reformulate the problem of text-image retrieval as that of learning a translation between the textual and visual domain. Our proposal leverages an end-to-end trainable architecture that can translate text into image features and vice versa and regularizes this mapping with a cycle-consistency criterion. Experimental evaluations for text-to-image and image-to-text retrieval, conducted on small, medium and large-scale datasets show consistent improvements over the baselines, thus confirming the appropriateness of using a cycle-consistent constrain for the text-image matching task.