-
Notifications
You must be signed in to change notification settings - Fork 3
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
3 changed files
with
132 additions
and
0 deletions.
There are no files selected for viewing
Empty file.
Empty file.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,132 @@ | ||
| <!DOCTYPE html> | ||
| <html> | ||
|
|
||
| <head> | ||
| <meta charset="utf-8"> | ||
| <meta name="viewport" content="width=device-width, initial-scale=1"> | ||
| <meta http-equiv="Content-Type" content="text/html; charset=UTF-8"> | ||
| <title>MMtrack</title> | ||
| <meta content="A robust single object tracker in LiDAR scenes following the motion-centric paradigm" name="description"> | ||
| <meta content="Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds" property="og:title"> | ||
| <meta content="A robust single object tracker in LiDAR scenes following the motion-centric paradigm." property="og:description"> | ||
| <meta content="http://people.eecs.berkeley.edu/~tancik/nerf/website_renders/images/nerf_graph.jpg" property="og:image"> | ||
| <meta content="Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds" property="twitter:title"> | ||
| <meta content="A robust single object tracker in LiDAR scenes following the motion-centric paradigm." property="twitter:description"> | ||
| <meta content="http://people.eecs.berkeley.edu/~tancik/nerf/website_renders/images/nerf_graph.jpg" property="twitter:image"> | ||
| <meta property="og:type" content="website"> | ||
| <meta content="summary_large_image" name="twitter:card"> | ||
| <meta content="width=device-width, initial-scale=1" name="viewport"> | ||
| <link href="./assets/css/bootstrap.min.css" rel="stylesheet" integrity="sha384-1BmE4kWBq78iYhFldvKuhfTAU6auU8tT94WrHftjDbrCEXSU1oBoqyl2QvZ6jIW3" crossorigin="anonymous"> | ||
| <link href="./assets/css/main.css" rel="stylesheet"> | ||
| </head> | ||
|
|
||
| <body class="container-sm"> | ||
| <div> | ||
| <h2 class="text-center d-block text-dark pt-5"> | ||
| Beyond 3D Siamese Tracking: <br>A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds | ||
| </h2> | ||
| <h4 class="text-center d-block" style="color:#ffc107">CVPR 2022 (Oral) | ||
| </h4> | ||
| <p class="text-center d-block text-dark"><a class="text-secondary" href="https://github.com/Ghostish/">Chaoda Zheng,</a> <a class="text-secondary" href="https://yanx27.github.io/">Xu Yan,</a> <a class="text-secondary" href="https://github.com/zhanghm1995">Haiming Zhang,</a> Baoyuan Wang, | ||
| Shenghui Cheng, Shuguang Cui, <a class="text-secondary" href="https://mypage.cuhk.edu.cn/academics/lizhen/"> Zhen Li</a></p> | ||
| <p class="text-center d-block text-dark">The Chinese University of Hong Kong, Shenzhen </p> | ||
|
|
||
| <div> | ||
| <div class="row gx-5 justify-content-center row-cols-2"> | ||
| <div class="col-1 p-3 text-decoration-none mx-5"> | ||
| <button type="button" class="btn"><a class="d-block mx-auto" href="https://arxiv.org/abs/2203.01730"><img src="./figs/paper_icon.png" width="100%"/> | ||
| </a>Paper</button> | ||
|
|
||
| </div> | ||
| <div class="col-1 p-3 text-decoration-none mx-5"> | ||
| <button type="button" class="btn"><a class="d-block mx-auto" href="https://github.com/Ghostish/Open3DSOT"><img src="./figs/code_icon.png" width="100%"/> | ||
| </a>Code</button> | ||
| </div> | ||
| </div> | ||
| </div> | ||
| </div> | ||
|
|
||
| <video class="w-50 mx-auto d-block mb-4" controls> | ||
| <source src="https://github.com/Ghostish/MM-Track/raw/main/figs/video.MOV" type="video/mp4"> | ||
| Your browser does not support the video tag. | ||
| </video> | ||
|
|
||
|
|
||
|
|
||
| <div class="bg-light container-sm p-4 w-75"> | ||
| <h3 class="subtitle">Motivation & Method</h3> | ||
| <p> | ||
| For single object tracking in LiDAR scenes (LiDAR SOT), previous methods rely on appearance matching to localize the target using a target template. | ||
| </p> | ||
| <img src="./figs/matching_paradigm.png" class="img_fluid d-block mx-auto w-100" alt="Siamese matching paradigm" /> | ||
| <p> | ||
| However, as shown in the following figure, matching-based approaches become unreliable when dealing with drastic appearance changes and distractors, which commonly exist in LiDAR scenes. | ||
| </p> | ||
| <img src="./figs/demo0.png" class="img_fluid d-block mx-auto w-100" alt="Distracted cases" /> | ||
| <p> | ||
| Since the task deals with a dynamic scene across a video sequence, the target's movements among successive frames provide useful cues to distinguish distractors and handle appearance changes. We for the first time present a <b>motion-centric paradigm</b> to handle LiDAR SOT. By explicitly learning from various "relative target motions" in data, the paradigm robustly localize the target in the current frame via motion transformation. | ||
| </p> | ||
| <img src="./figs/motion_centric_paradigm.png" class="img_fluid d-block mx-auto w-100" alt="Motion-Centric paradigm" /> | ||
| <p> | ||
| Based on the motion-centric paradigm, a two-stage tracker M^2-Track is proposed. At 1 st-stage, M^2-Track localizes the target within successive frames via motion transformation. Then it refines the target box through motion-assisted shape completion | ||
| at 2nd-stage. M^2-Track significantly outperforms the previous SOTAs and further shows its potential when simply integrated with appearance-matching. | ||
| </p> | ||
| <img src="./figs/arch.png" class="img_fluid d-block mx-auto w-100" alt="M^2-Track Architecture" /> | ||
| </div> | ||
|
|
||
| <div class="container-sm p-4 w-75"> | ||
| <h3 class="subtitle">Distractor Statistics</h3> | ||
| <p> Distributions of distractors for car/vehicle objects on different datasets: | ||
| </p> | ||
| <div class="row mx-auto"> | ||
| <img src="./figs/distractor_statistics.png" class=" d-block mx-auto" alt="distractor statistics" /> | ||
| </div> | ||
| <p>Visualization:</p> | ||
| <div class="row mx-auto"> | ||
| <img src="./figs/distractors_vis.png" class=" d-block mx-auto" alt="distractor statistics" /> | ||
| </div> | ||
| <p>NuScenes and Waymo are more challenging for matching-based approaches due to widespread distractors in scenes. But M^2-Track robustly handles distractors via explicit motion modeling.</p> | ||
|
|
||
| </div> | ||
|
|
||
| <div class="container-sm p-4 w-75"> | ||
| <h3 class="subtitle">Quantitative Results</h3> | ||
| <h6 class="text-black p-1">NuScenes & Waymo</h6> | ||
| <div class="row mx-auto"> | ||
| <img src="./figs/results_nusc_waymo.png" class=" d-block mx-auto" alt="results on nuscenes and waymo" /> | ||
| </div> | ||
| <h6 class="text-black p-3">Comparison & Behavior Analysis in KITTI </h6> | ||
| <div class="row mx-auto justify-content-center"> | ||
| <img src="./figs/results_kitti.png" class="d-block col-6" alt="results on KITTI" /> | ||
| <div class="col-6"> | ||
| <img src="./figs/distractors.png" class="d-block w-100" alt="robustness to distractors" /> | ||
| <img src="./figs/with_appearance_matching.png" class="d-block w-100 my-auto" alt="robustness to distractors" /> | ||
| </div> | ||
|
|
||
|
|
||
| </div> | ||
|
|
||
| </div> | ||
| <div class="container-sm p-4 w-75"> | ||
| <h3 class="subtitle">Qualitative Results</h3> | ||
| <h6 class="text-black p-1">Tracking on Cars</h6> | ||
|
|
||
| <img src="./figs/vis2.png" class="img-fluid d-block mx-auto" alt="results on nuscenes and waymo" /> | ||
| <h6 class="text-black p-1">Tracking on Pedestrian</h6> | ||
| <img src="./figs/vis3.png" class="img-fluid d-block mx-auto" alt="results on nuscenes and waymo" /> | ||
| </div> | ||
| <div class="container-sm bg-light p-4 w-75"> | ||
| <h3 class="subtitle">Citation</h3> | ||
| <pre><code> | ||
| @article{zheng2022beyond, | ||
| title={Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds}, | ||
| author={Zheng, Chaoda and Yan, Xu and Zhang, Haiming and Wang, Baoyuan and Cheng, Shenghui and Cui, Shuguang and Li, Zhen}, | ||
| journal={arXiv preprint arXiv:2203.01730}, | ||
| year={2022} | ||
| } | ||
| </code></pre> | ||
| </div> | ||
|
|
||
| </body> | ||
|
|
||
| </html> |