Symmetry and Uncertainty-Aware Object SLAM for 6DoF Object Pose Estimation

Related tags

Deep Learningsuo_slam
Overview

SUO-SLAM

This repository hosts the code for our CVPR 2022 paper "Symmetry and Uncertainty-Aware Object SLAM for 6DoF Object Pose Estimation". ArXiv link.

Citation

If you use any part of this repository in an academic work, please cite our paper as:

@inproceedings{Merrill2022CVPR,
  Title      = {Symmetry and Uncertainty-Aware Object SLAM for 6DoF Object Pose Estimation},
  Author     = {Nathaniel Merrill and Yuliang Guo and Xingxing Zuo and Xinyu Huang and Stefan Leutenegger and Xi Peng and Liu Ren and Guoquan Huang},
  Booktitle  = {2022 Conference on Computer Vision and Pattern Recognition (CVPR)},
  Year       = {2022},
  Address    = {New Orleans, USA},
  Month      = jun,
}

Installation

Click for details... This codebase was tested on Ubuntu 18.04. To use the BOP rendering (i.e. for keypoint labeling) install
sudo apt install libfreetype6-dev libglfw3

You will also need a python environment that contains the required packages. To see what packages we used, check out the list of requirements in requirements.txt. They can be installed via pip install -r requirements.txt

Preparing Data

Click for details...

Datasets

To be able to run the training and testing (i.e. single view or with SLAM), first decide on a place to download the data to. The disk will need a few hundred GB of space for all the data (at least 150GB for download and more to extract). All of our code expects the data to be in a local directory ./data, but you can of course symlink this to another location (perhaps with more disk space). So, first of all, in the root of this repo run

$ mkdir data

or to symlink to an external location

$ ln -s /path/to/drive/with/space/ ./data

You can pick and choose what data you want to download (for example just T-LESS or YCBV). Note that all YCBV and TLESS downloads have our keypoint labels packaged along with the data. Download the following google drive links into ./data and extract them.

When all is said and done, the tree should look like this

$ cd ./data && tree --filelimit 3
.
├── bop_datasets
│   ├── tless 
│   └── ycbv 
├── saved_detections
└── VOCdevkit
    └── VOC2012

Pre-trained models

You can download the pretrained models anywhere, but I like to keep them in the results directory that is written to during training.

Training

Click for details...

First set the default arguments in ./lib/args.py for your username if desired, then execute

$ ./train.py

with the appropriate arguments for your filesystem. You can also run

$ ./train.py -h

for a full list of arguments and their meaning. Some important args are batch_size, which is the number of images loaded for each training batch. Note that there may be a variable number of objects in each image, and the objects are all stacked together into one big batch to run the network -- so the actual batch size being run might be multiple times batch_size. In order to keep batch_size reasonably large, we provide another arg called truncate_obj, which, as the help says, truncates the object batches to this number if it exceeds it. We recommend that you start with a large batch size so that you can find out the maximum truncate_obj for you GPUs, then reduce the batch size until there are little to no warnings about too many objects being truncated.

Evaluation

Click for details...

Before you can evaluate in a single-view or SLAM fashion, you will need to build the thirdparty libraries for PnP and graph optimization. First make sure that you have CERES solver installed. The run

$ ./build_thirdparty.sh

Reproducing Results

To reproduce the results of the paper with the pretrained models, check out the scripts under the scripts directory:

eval_all_tless.sh  eval_all_ycbv.sh  make_video.sh

These will reproduce most of the results in the paper as well as any video clips you want. You may have to change the first few lines of each script. Note that these examples can also show you the proper arguments if you want to run from command line alone.

Note that for the T-LESS dataset, we use the thirdparty BOP toolkit to get the VSD error recall, which will show up in the final terminal output as "Mean object recall" among other numbers.

Labeling

Click for details...

Overview

We manually label keypoints on the CAD model to enable some keypoints with semantic meaning. For the full list of keypoint meanings, see the specific README

We provide our landmark labeling tool. Check out the script manual_keypoints.py. This same script can be used to make a visualization of the keypoints as shown below with the --viz option.

The script will show a panel of the same object but oriented slightly differently. The idea is that you pick the same keypoint multiple times to ensure correctness and to get a better label by averaging multiple samples.

The script will also print the following directions to follow in the terminal.

============= Welcome ===============
Select the keypoints with a left click!
Use the "wasd" to turn the objects.
Press "i" to zoom in and "o" to zoom out.
Make sure that the keypoint colors match between all views.
Messed up? Just press 'u' to undo.
Press "Enter" to finish and save the keypoints
Press "Esc" to just quit

Once you have pressed "enter", you will get to an inspection pane.

Where the unscaled mean keypoints are on the left, and the ones scaled by covariance is on the left, where the ellipses are the Gaussian 3-sigma projected onto the image. If the covariance is too large, or the mean is out of place, then you may have messed up. Again, the program will print out these directions to terminal:

Inspect the results!
Use the "wasd" to turn the object.
Press "i" to zoom in and "o" to zoom out.
Press "Esc" to go back, "Enter" to accept (saving keypoints and viewpoint for vizualization).
Please pick a point on the object!

So if you are done, and the result looks good, then press "Enter", if not then "Esc" to go back. Make sure also that when you are done, you rotate and scale the object into the best "view pose" (with the front facing the camera, and top facing up), as this pose is used by both the above vizualization and the actual training code for determining the best symmetry to pick for an initial detection.

Labeling Tips

Even though there are 8 panels, you don't need to fill out all 8. Each keypoint just needs at least 3 samples to sample the covariance.

We recommend that you label the same keypoint (say keypoint i) on all the object renderings first, then go to the inspection panel at the end of this each time so that you can easily undo a mistake for keypoint i with the "u" key and not lose any work. Otherwise, if you label each object rendering completely, then you may have to undo a lot of labelings that were not mistakes.

Also, if there is an object that you want to label a void in the CAD model, like the top center of the bowl, then you can use the multiple samples to your advantage, and choose samples that will average to the desired result, since the labels are required to land on the actual CAD model in the labeling tool.

<\details>

Owner
Robot Perception & Navigation Group (RPNG)
Research on robot sensing, estimation, localization, mapping, perception, and planning
Robot Perception & Navigation Group (RPNG)
This is a JAX implementation of Neural Radiance Fields for learning purposes.

learn-nerf This is a JAX implementation of Neural Radiance Fields for learning purposes. I've been curious about NeRF and its follow-up work for a whi

Alex Nichol 62 Dec 20, 2022
This repository contains the code for the paper Neural RGB-D Surface Reconstruction

Neural RGB-D Surface Reconstruction Paper | Project Page | Video Neural RGB-D Surface Reconstruction Dejan Azinović, Ricardo Martin-Brualla, Dan B Gol

Dejan 406 Jan 04, 2023
Learning Generative Models of Textured 3D Meshes from Real-World Images, ICCV 2021

Learning Generative Models of Textured 3D Meshes from Real-World Images This is the reference implementation of "Learning Generative Models of Texture

Dario Pavllo 115 Jan 07, 2023
1st Place Solution to ECCV-TAO-2020: Detect and Represent Any Object for Tracking

Instead, two models for appearance modeling are included, together with the open-source BAGS model and the full set of code for inference. With this code, you can achieve around 79 Oct 08, 2022

A pytorch reproduction of { Co-occurrence Feature Learning from Skeleton Data for Action Recognition and Detection with Hierarchical Aggregation }.

A PyTorch Reproduction of HCN Co-occurrence Feature Learning from Skeleton Data for Action Recognition and Detection with Hierarchical Aggregation. Ch

Guyue Hu 210 Dec 31, 2022
Semantic Segmentation for Aerial Imagery using Convolutional Neural Network

This repo has been deprecated because whole things are re-implemented by using Chainer and I did refactoring for many codes. So please check this newe

Shunta Saito 27 Sep 23, 2022
CRISCE: Automatically Generating Critical Driving Scenarios From Car Accident Sketches

CRISCE: Automatically Generating Critical Driving Scenarios From Car Accident Sketches This document describes how to install and use CRISCE (CRItical

Chair of Software Engineering II, Uni Passau 2 Feb 09, 2022
BoxInst: High-Performance Instance Segmentation with Box Annotations

Introduction This repository is the code that needs to be submitted for OpenMMLab Algorithm Ecological Challenge, the paper is BoxInst: High-Performan

88 Dec 21, 2022
Python library for tracking human heads with FLAME (a 3D morphable head model)

Video Head Tracker 3D tracking library for human heads based on FLAME (a 3D morphable head model). The tracking algorithm is inspired by face2face. It

61 Dec 25, 2022
Using Convolutional Neural Networks (CNN) for Semantic Segmentation of Breast Cancer Lesions (BRCA)

Using Convolutional Neural Networks (CNN) for Semantic Segmentation of Breast Cancer Lesions (BRCA). Master's thesis documents. Bibliography, experiments and reports.

Erick Cobos 73 Dec 04, 2022
Implementation of momentum^2 teacher

Momentum^2 Teacher: Momentum Teacher with Momentum Statistics for Self-Supervised Learning Requirements All experiments are done with python3.6, torch

jemmy li 121 Sep 26, 2022
Software & Hardware to do multi color printing with Sharpies

3D Print Colorizer is a combination of 3D printed parts and a Cura plugin which allows anyone with an Ender 3 like 3D printer to produce multi colored

343 Jan 06, 2023
CompilerGym is a library of easy to use and performant reinforcement learning environments for compiler tasks

CompilerGym is a library of easy to use and performant reinforcement learning environments for compiler tasks

Facebook Research 721 Jan 03, 2023
Scripts for training an AI to play the endless runner Subway Surfers using a supervised machine learning approach by imitation and a convolutional neural network (CNN) for image classification

About subwAI subwAI - a project for training an AI to play the endless runner Subway Surfers using a supervised machine learning approach by imitation

82 Jan 01, 2023
BossNAS: Exploring Hybrid CNN-transformers with Block-wisely Self-supervised Neural Architecture Search

BossNAS This repository contains PyTorch evaluation code, retraining code and pretrained models of our paper: BossNAS: Exploring Hybrid CNN-transforme

Changlin Li 127 Dec 26, 2022
GARCH and Multivariate LSTM forecasting models for Bitcoin realized volatility with potential applications in crypto options trading, hedging, portfolio management, and risk management

Bitcoin Realized Volatility Forecasting with GARCH and Multivariate LSTM Author: Chi Bui This Repository Repository Directory ├── README.md

Chi Bui 113 Dec 29, 2022
Code for ECCV 2020 paper "Contacts and Human Dynamics from Monocular Video".

Contact and Human Dynamics from Monocular Video This is the official implementation for the ECCV 2020 spotlight paper by Davis Rempe, Leonidas J. Guib

Davis Rempe 207 Jan 05, 2023
ICNet for Real-Time Semantic Segmentation on High-Resolution Images, ECCV2018

ICNet for Real-Time Semantic Segmentation on High-Resolution Images by Hengshuang Zhao, Xiaojuan Qi, Xiaoyong Shen, Jianping Shi, Jiaya Jia, details a

Hengshuang Zhao 594 Dec 31, 2022
Myia prototyping

Myia Myia is a new differentiable programming language. It aims to support large scale high performance computations (e.g. linear algebra) and their g

Mila 456 Nov 07, 2022
Beyond Image to Depth: Improving Depth Prediction using Echoes (CVPR 2021)

Beyond Image to Depth: Improving Depth Prediction using Echoes (CVPR 2021) Kranti Kumar Parida, Siddharth Srivastava, Gaurav Sharma. We address the pr

Kranti Kumar Parida 33 Jun 27, 2022