Douglas Meneghetti

This paper proposes a novel algorithm, called CAPTION, for identifying and correcting errors in automatically generated image captions. The algorithm combines Deep Learning (DL) for object detection in images with Natural Language Processing techniques. CAPTION has been tested in the following three tasks: (1) classify a caption as correct or not; (2) detect wrong words in the caption, and (3) suggest text corrections. Results show that our method is superior with respect to others evaluated in the same data set in the error correction task. These other methods are generally based exclusively on DL models. This work shows that, although semantics still has not been used at its fullest in this type of task, a combination of DL with Natural Language Processing tools presents a better overall performance than using DL methods alone.

This paper presents catsim, the first package written in the Python language specialized in computerized adaptive tests and the logistical models of Item Response Theory. catsim provides functions for generating item and examinee parameters, simulating tests and plotting results, as well as enabling end users to create new procedures for proficiency initialization, item selection, proficiency estimation and test stopping criteria. The simulator keeps a record of the items selected for each examinee as well as their answers and also enables the simulation of linear tests, in which all examinees answer the same items. The various components made available by catsim can also be used in the creation of third-party testing applications. Examples of such usages are also presented in this paper.

Annotated image dataset of household objects from the RoboFEI@Home teamThis data set contains two sets of pictures of household objects, created by the RoboFEI@Home team to develop object detection systems for a domestic robot.The first data set was created with objects from a local supermarket. Product brands are typical from Brazil. The second data set is composed of objects from the RoboCup@Home 2018 OPL competition.

Object detection techniques that achieve state-of-the-art detection accuracy employ convolutional neural networks, implemented to have optimal performance in graphics processing units. Some hardware systems, such as mobile robots, operate under constrained hardware situations, but still benefit from object detection capabilities. Multiple network models have been proposed, achieving comparable accuracy with reduced architectures and leaner operations. Motivated by the need to create an object detection system for a soccer team of mobile robots, this work provides a comparative study of recent proposals of neural networks targeted towards constrained hardware environments, in the specific task of soccer ball detection. We train multiple open implementations of MobileNetV2 and MobileNetV3 models with different underlying architectures, as well as YOLOv3, TinyYOLOv3, YOLOv4 and TinyYOLOv4 in an annotated image data set captured using a mobile robot. We then report their mean average precision on a test data set and their inference times in videos of different resolutions, under constrained and unconstrained hardware configurations. Results show that MobileNetV3 models have a good trade-off between mAP and inference time in constrained scenarios only, while MobileNetV2 with high width multipliers are appropriate for server-side inference. YOLO models in their official implementations are not suitable for inference in CPUs.