Submission: Félix&Neves-ICRA2017-IET2019/ITODD

Download submission
Submission name
Submission time (UTC) Oct. 22, 2019, 5:31 a.m.
User inesdfelix
Task Model-based 6D localization of seen objects
Dataset ITODD
Training model type Default
Training image type Synthetic (custom)
Description
Evaluation scores
AR:0.069
AR_MSPD:0.073
AR_MSSD:0.084
AR_VSD:0.050
average_time_per_image:35.487

Method: Félix&Neves-ICRA2017-IET2019

User inesdfelix
Publication [1] Rodrigues, P., Antunes, M., Raposo, C., Marques, P., Fonseca, F., Barreto, J. (2019). Deep segmentation leverages geometric pose estimation in computer-aided total knee arthroplasty. Healthcare Technology Letters. 10.1049/htl.2019.0078.; [2] Raposo, C., Barreto, J.P.: Using 2 point+normal sets for fast registration of point clouds with small overlap. In: 2017 IEEE International Conference on Robotics and Automation (ICRA). pp. 5652–5658 (May 2017)
Implementation
Training image modalities RGB-D
Test image modalities RGB-D
Description

Implementation of a similar method of [1], but segmentation was done using Mask R-CNN (instead of U-Net) and trained a specific network to each dataset. For training, we used real training images, when provided in the dataset. In the remaining datasets, we generated synthetic images with multiple objects/ instances of objects with random backgrounds. The pose estimation method is the one presented in [2], refined using a standard ICP approach. Implemented by Inês Félix and Miguel Neves, in collaboration with Michel Antunes, Pedro Rodrigues and Carolina Raposo.

Computer specifications