Region Growing Refinement (RGR)
Our unsupervised algorithm for refinement of semantic segmentation masks. Available in MATLAB and Python.
DIAS, P. A.; MEDEIROS, H. “Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections,” Asian Conference on Computer VIsion (ACCV), 2018.
FreeLabel: A Publicly Available Annotation Tool based on Freehand Traces
Our open-source web-based user interface for annotation of segmentation datasets. Using our unsupervised RGR algorithm for label propagation, the interface can be easily adapted to any dataset. Implementation in Python+HTML, using Django.
DIAS, P. A.; TABB, A.; MEDEIROS, H. “FreeLabel: A Publicly Available Annotation Tool based on Freehand Traces,” Winter Conference on Applications of Computer Vision (WACV), 2019.
Multi-species Fruit Flower Segmentation Network
Our finetune DeeplabV2+RGR pipeline for segmentation of fruit flowers. Trained on images of apple flowers, the available model generalizes well to images containing pears and peaches flowers, in a variety of acquisition conditions
DIAS, P. A.; TABB, A.; MEDEIROS, H. “Multispecies Fruit Flower Detection Using a Refined Semantic Segmentation Network” IEEE Robotics and Automation Letters, vol. 3, no. 4, 2018.
Gaze Estimation for Assisted Living Environments
Project in collaboration with the University of Genoa (DIBRIS/MaLGa)
Our gaze regression network that relies only on the relative positions of 5 facial keypoints, detected using the off-the-shelf OpenPose model. This work introduces the concept of Confidence Gated Units (CGU) to handle cases of missing/poorly detected input features. Plus, our model provides uncertainty estimations of its own gaze predictions.
DIAS, P. A.; MALAFRONTE, D; MEDEIROS, H. & ODONE, F. (2019). Gaze Estimation for Assisted Living Environments. Accepted for WACV 2020. Preprint available at arXiv preprint arXiv:1909.09225.