sollynoay / a2fnet

A2FNet

This repository contains the source code for our paper:
Elevation Angle Estimation in 2D Acoustic Images Using Pseudo Front View
RAL and ICRA2021

Libraries

pytorch 1.2.0
torchvision 0.40
tensorboardX

Train

python train.py --train_path to_your_train_label --test_path to_your_validate_label

During training, using front depth map instead of inverse front depth map can achieve better results

Dataset

The simulation dataset used in this paper can be download from the following links.
water tank
floating object

The structure of the dataset is as follows.

├── water-tank
│   ├── bricks
        ├── sfront.txt //inverse front depth map without quantization loss
        ├── resizeimg.png // synthetic acoustic image
        ├── ele_resize.png // ground truth of elevation angle
        ├── front_resize.png // inverse front depth map for visualization only
│   └── cylinder

Simulator

For the simulator used to generate synthetic datasets, check here.

Evaluation

How to evaluate depth map is shown in eval.ipynb.

a2fnet's Issues

Hello, I want to ask a question about the network

Thank you very much for sharing, your code has provided a great help for my academic research.

Based on my current understanding of the code, does the part of the code you upload only generate a pseudo front view?
Follow-up technology----use a 3D CNN decoder to generate the 3D shape, have you not uploaded it on GITHUB？

I hope you can answer my doubts. This is very helpful for me as a beginner in 3D sonar reconstruction.

Thank you again for sharing.