├── LICENSE
├── README.md
├── figs
├── git_compare.png
├── git_vis.png
└── git_vis2.png
└── gifs
├── 1.gif
└── 2.gif
/LICENSE:
--------------------------------------------------------------------------------
1 | MIT License
2 |
3 | Copyright (c) 2024 RascalGdd
4 |
5 | Permission is hereby granted, free of charge, to any person obtaining a copy
6 | of this software and associated documentation files (the "Software"), to deal
7 | in the Software without restriction, including without limitation the rights
8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 |
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 |
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Vanishing-Point-Guided Video Semantic Segmentation of Driving Scenes (CVPR2024 Highlight)
2 | > [Paper](https://arxiv.org/pdf/2401.15261.pdf)
3 | >
4 | > Authors:
5 | > [Diandian Guo](https://scholar.google.com/citations?user=yXycwhIAAAAJ&hl=en),
6 | > [Deng-Ping Fan](https://dengpingfan.github.io/),
7 | > [Tongyu Lu](https://lucainiaoge.github.io/),
8 | > [Christos Sakaridis](https://people.ee.ethz.ch/~csakarid/),
9 | > [Luc Van Gool](https://ee.ethz.ch/the-department/faculty/professors/person-detail.OTAyMzM=.TGlzdC80MTEsMTA1ODA0MjU5.html).
10 | >
11 | ## 1. Vanishing-Point Guided
12 | The estimation of implicit cross-frame correspondences and the high computational cost have long been major challenges in video semantic segmentation (VSS) for driving scenes.
13 | Prior works utilize keyframes, feature propagation, or cross-frame attention to address these issues.
14 | By contrast, we are the first to harness vanishing point (VP) priors for more effective segmentation. The comparisons between different SOTA methods are shown in the following figure.
15 |
16 |
17 |
18 | Figure 1: Qualitative comparison on ACDC. The yellow box represents the densely partitioned VP region. Our model produces more accurate results for both distant tiny hard samples near the VP and occluded fast-moving close targets.
19 |
20 |
21 |
22 | | *VPSeg-Sample_1* | *VPSeg-Sample_2* |
23 | | :------------------------------: | :-------------------------------: |
24 | |
|
|
25 |
26 |
27 | ## 2. Visualization
28 | Visualization of detail attention maps O with N motion attention layers in CMA. As N increases, the detail attention map interacts more heavily with the dynamic features, and the weights gradually decrease in closer parts of the scene or on simple semantic categories. The highlighted distant regions near the VP suggest that the final predictions P_f are primarily based on the detail-based predictions P_d and not on P_c for these regions. The VP proximity map serves as a positional prior and assists the model in pinpointing the locations of these distant regions.
29 |
30 |
31 |
32 |
33 | ## 3. Quantitative Results
34 | The results for various methods are shown below.
35 | | Method | Publication | Backbone | Params(M)| mIoU (A.)| mIoU (C.) | miIoU(A.)| miIoU(C.)| mIA-IoU | VSS |
36 | |------------------|:-----------:|:--------:|:--------:|:--------:|:---------:|:--------:|:--------:|:--------:|:--------:|
37 | | DeepLabv3+ | ECCV'18 | ResNet-101 | 62.7 | 72.79 | 79.09 | 43.14 | 56.89 | 36.32 | ✗ |
38 | | PSPNet | CVPR'17 | ResNet-101 | 68.0 | 72.26 | 78.34 | 40.95 | 56.78 | 37.29 | ✗ |
39 | | OCRNet | ECCV'20 | ResNet-101 | 55.6 | 70.39 | 80.09 | 43.36 | 59.55 | 33.38 | ✗ |
40 | | SeaFormer | ICLR'23 | SeaFormer-L | 14.0 | 70.09 | 77.70 | 40.41 | 56.96 | 33.26 | ✗ |
41 | | SegFormer | NIPS'21 | MiT-B1 | 13.7 | 70.25 | 78.56 | 41.14 | 58.25 | 34.05 | ✗ |
42 | | SegFormer | NIPS'21 | MiT-B3 | 44.6 | 75.38 | 81.32 | 46.51 | 60.01 | 38.42 | ✗ |
43 | | Video K-Net | CVPR'22 | Swin-B | 104.6 | 69.03 | 76.62 | 39.33 | 56.02 | 31.67 | ✓ |
44 | | ETC | ECCV'20 | ResNet-101 | 68,1 | 71.45 | 79.50 | 42.28 | 58.35 | 36.71 | ✓ |
45 | | TCB | CVPR'21 | ResNet-101 | 72.5 | 70.56 | 78.70 | 41.76 | 57.84 | 35.96 | ✓ |
46 | | NetWarp | ICCV'17 | PSPNet | 90.6 | 73.71 | 80.60 | 45.59 | 59.63 | 36.58 | ✓ |
47 | | CFFM | CVPR'22 | MiT-B3 | 49.6 | 75.47 | 81.44 | 47.31 | 60.09 | 37.88 | ✓ |
48 | | MRCFA | ECCV'22 | MiT-B3 | 48.2 | 75.63 | 81.31 | 46.28 | 60.56 | 38.81 | ✓ |
49 | | VPSeg (Ours) | CVPR'24 | MiT-B1 | 14.9 | 72.86 | 79.56 | 43.42 | 59.53 | 37.96 | ✓ |
50 | | VPSeg (Ours) | CVPR'24 | MiT-B3 | 46.8 | 77.48 | 82.46 | 49.42 | 61.79 | 41.48 | ✓ |
51 |
52 | ## 4. Code
53 |
54 | As I have signed the non-disclosure agreement (NDA) with ETH and TOYOTA, I'm strictly forbidden to share the code.
55 |
56 | ## 5. Citations
57 | ```bibtex
58 | @inproceedings{guo2024vpseg,
59 | title={Vanishing-Point-Guided Video Semantic Segmentation of Driving Scenes},
60 | author={Diandian Guo and Fan, Deng-Ping and Tongyu Lu and Sakaridis, Christos and Van Gool, Luc},
61 | booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision and Patern Recognition (CVPR)},
62 | year={2024}
63 | }
64 | ```
65 | ## 6. Contact
66 | - Diandian Guo, guodiandian1998@gmail.com
67 |
--------------------------------------------------------------------------------
/figs/git_compare.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RascalGdd/VPSeg/37f0420ffcd16fe1f78372a9836395f32dcf0ee8/figs/git_compare.png
--------------------------------------------------------------------------------
/figs/git_vis.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RascalGdd/VPSeg/37f0420ffcd16fe1f78372a9836395f32dcf0ee8/figs/git_vis.png
--------------------------------------------------------------------------------
/figs/git_vis2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RascalGdd/VPSeg/37f0420ffcd16fe1f78372a9836395f32dcf0ee8/figs/git_vis2.png
--------------------------------------------------------------------------------
/gifs/1.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RascalGdd/VPSeg/37f0420ffcd16fe1f78372a9836395f32dcf0ee8/gifs/1.gif
--------------------------------------------------------------------------------
/gifs/2.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/RascalGdd/VPSeg/37f0420ffcd16fe1f78372a9836395f32dcf0ee8/gifs/2.gif
--------------------------------------------------------------------------------