Neural Scene Flow Prior (NeurIPS 2021 spotlight)

Last update: Jan 03, 2023

Overview

Neural Scene Flow Prior

Xueqian Li, Jhony Kaesemodel Pontes, Simon Lucey

Will appear on Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS, 2021) as spotlight.

arXiv link: https://arxiv.org/pdf/2111.01253.pdf

Scene flow results on Argoverse
Point cloud integration (left: densified point cloud; right: sparse point cloud)

Prerequisites

This code is based on PyTorch implementation, and tested on torch=1.6.0 with CUDA 10.1 OR torch=1.7.1 with CUDA 10.2.

For a detailed installation guide, please go to requirements.txt.

Dataset

We provide four datasets we used in our paper. You may download datasets used in the paper from these anonymous links:

KITTI (266MB)
Argoverse (5.3GB)
nuScenes (436MB)
FlyingThings3D (436MB)

After you download the dataset, you can create a symbolic link in the ./dataset folder as ./dataset/kitti, ./dataset/argoverse, ./dataset/nuscenes, and ./dataset/flyingthings.

Optimization

Since we use neural scene flow prior for runtime optimization, our method does not include any "training".

Just run following lines for a simple optimization on a small KITTI Scene Flow dataset (only 50 testing samples)

python optimization.py \
--dataset KITTISceneFlowDataset \
--dataset_path dataset/kitti \
--exp_name KITTI_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--num_points 2048 \
--hidden_units 128 \
--lr 0.008 \
--backward_flow \
--early_patience 70 \
--visualize

You can then play with these configurations. We provide commands we used to generate results in the small point coud (2048 points) experiments and large point cloud (all points included) experiments.

1. small point cloud (2048 points)

KITTI Scene Flow

python optimization.py \
--dataset KITTISceneFlowDataset \
--dataset_path dataset/kitti \
--exp_name KITTI_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--num_points 2048 \
--hidden_units 128 \
--lr 0.008 \
--backward_flow \
--early_patience 70 \
--visualize

Argoverse Scene Flow

python optimization.py \
--dataset ArgoverseSceneFlowDataset \
--dataset_path dataset/argoverse \
--exp_name Argoverse_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--num_points 2048 \
--hidden_units 128 \
--lr 0.008 \
--backward_flow \
--early_patience 30 \
--visualize

nuScenes Scene Flow

python optimization.py \
--dataset NuScenesSceneFlowDataset \
--dataset_path dataset/nuscenes \
--exp_name Argoverse_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--num_points 2048 \
--hidden_units 128 \
--lr 0.008 \
--backward_flow \
--early_patience 30 \
--visualize

FlyingThings3D

python optimization.py \
--dataset FlyingThings3D \
--dataset_path dataset/flyingthings \
--exp_name FlyingThings_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--num_points 2048 \
--hidden_units 128 \
--lr 0.008 \
--backward_flow \
--early_patience 30 \
--visualize

2. dense point cloud (all points included)

KITTI Scene Flow

python optimization.py \
--dataset KITTISceneFlowDataset \
--dataset_path dataset/kitti \
--exp_name KITTI_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--use_all_points \
--hidden_units 128 \
--lr 0.001 \
--early_patience 100 \
--visualize

Argoverse Scene Flow

python optimization.py \
--dataset ArgoverseSceneFlowDataset \
--dataset_path dataset/argoverse \
--exp_name Argoverse_2048_points \
--batch_size 1 \
--iters 5000 \
--compute_metrics \
--use_all_points \
--hidden_units 128 \
--lr 0.003 \
--backward_flow \
--early_patience 100 \
--visualize

Contributing

If you find the project useful for your research, you may cite,

@article{li2021neural,
  title={Neural scene flow prior},
  author={Li, Xueqian and Pontes, Jhony Kaesemodel and Lucey, Simon},
  journal={arXiv preprint arXiv:2111.01253},
  year={2021}
}

Neural Scene Flow Prior (NeurIPS 2021 spotlight)

Related tags

Overview

Neural Scene Flow Prior

Scene flow results on Argoverse

Point cloud integration (left: densified point cloud; right: sparse point cloud)

Prerequisites

Dataset

Optimization

1. small point cloud (2048 points)

KITTI Scene Flow

Argoverse Scene Flow

nuScenes Scene Flow

FlyingThings3D

2. dense point cloud (all points included)

KITTI Scene Flow

Argoverse Scene Flow

Contributing

Owner

Lilac Lee

Calibrated Hyperspectral Image Reconstruction via Graph-based Self-Tuning Network.

Pytorch Code for "Medical Transformer: Gated Axial-Attention for Medical Image Segmentation"

A highly modular PyTorch framework with a focus on Neural Architecture Search (NAS).

Pytorch version of VidLanKD: Improving Language Understanding viaVideo-Distilled Knowledge Transfer

PyTorch implementation of DCT fast weight RNNs

BEGAN in PyTorch

High-level library to help with training and evaluating neural networks in PyTorch flexibly and transparently.

Paddle Graph Learning (PGL) is an efficient and flexible graph learning framework based on PaddlePaddle

Cockpit is a visual and statistical debugger specifically designed for deep learning.

"NAS-Bench-301 and the Case for Surrogate Benchmarks for Neural Architecture Search".

This is the official implement of paper "ActionCLIP: A New Paradigm for Action Recognition"

CPT: A Pre-Trained Unbalanced Transformer for Both Chinese Language Understanding and Generation

Official PyTorch implementation of "Camera Distance-aware Top-down Approach for 3D Multi-person Pose Estimation from a Single RGB Image", ICCV 2019

Code for classifying international patents based on the text of their titles/abstracts

KDD CUP 2020 Automatic Graph Representation Learning: 1st Place Solution

Implementation of ViViT: A Video Vision Transformer

A Streamlit demo demonstrating the Deep Dream technique. Adapted from the TensorFlow Deep Dream tutorial.

Using Language Model to Bootstrap Human Activity Recognition Ambient Sensors Based in Smart Homes

Collection of generative models in Tensorflow

3D position tracking for soccer players with multi-camera videos