-
-
Notifications
You must be signed in to change notification settings - Fork 108
/
Copy pathREADME.md
47 lines (32 loc) · 7.33 KB
/
README.md
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
## Introduction
<a href="https://github.com/facebookresearch/Mask2Former">Official Repo</a>
<a href="https://github.com/SegmentationBLWX/sssegmentation/blob/main/ssseg/modules/models/segmentors/mask2former/mask2former.py">Code Snippet</a>
<details>
<summary align="left"><a href="https://arxiv.org/pdf/2112.01527.pdf">Mask2Former (CVPR'2022)</a></summary>
```latex
@inproceedings{cheng2021mask2former,
title={Masked-attention Mask Transformer for Universal Image Segmentation},
author={Bowen Cheng and Ishan Misra and Alexander G. Schwing and Alexander Kirillov and Rohit Girdhar},
journal={CVPR},
year={2022}
}
```
</details>
## Results
#### ADE20k
| Segmentor | Pretrain | Backbone | Crop Size | Schedule | Train/Eval Set | mIoU | Download |
| :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: |
| Mask2Former | ImageNet-1k-224x224 | Swin-T | 512x512 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/130 | train/val | 49.20% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swintiny_ade20k.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swintiny_ade20k.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swintiny_ade20k.log) |
| Mask2Former | ImageNet-1k-224x224 | Swin-S | 512x512 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/130 | train/val | 51.37% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swinsmall_ade20k.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinsmall_ade20k.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinsmall_ade20k.log) |
| Mask2Former | ImageNet-22k-384x384 | Swin-B | 640x640 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/130 | train/val | 54.13% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swinbase_ade20k.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinbase_ade20k.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinbase_ade20k.log) |
| Mask2Former | ImageNet-22k-384x384 | Swin-L | 640x640 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/130 | train/val | 56.30% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swinlarge_ade20k.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinlarge_ade20k.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinlarge_ade20k.log) |
#### Cityscapes
| Segmentor | Pretrain | Backbone | Crop Size | Schedule | Train/Eval Set | mIoU | Download |
| :-: | :-: | :-: | :-: | :-: | :-: | :-: | :-: |
| Mask2Former | ImageNet-1k-224x224 | Swin-T | 512x1024 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/500 | train/val | 82.10% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swintiny_cityscapes.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swintiny_cityscapes.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swintiny_cityscapes.log) |
| Mask2Former | ImageNet-1k-224x224 | Swin-S | 512x1024 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/500 | train/val | 82.65% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swinsmall_cityscapes.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinsmall_cityscapes.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinsmall_cityscapes.log) |
| Mask2Former | ImageNet-22k-384x384 | Swin-B | 512x1024 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/500 | train/val | 83.62% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swinbase_cityscapes.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinbase_cityscapes.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinbase_cityscapes.log) |
| Mask2Former | ImageNet-22k-384x384 | Swin-L | 512x1024 | LR/POLICY/BS/EPOCH: 0.0001/poly/16/500 | train/val | 83.79% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/mask2former/mask2former_swinlarge_cityscapes.py) | [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinlarge_cityscapes.pth) | [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_mask2former/mask2former_swinlarge_cityscapes.log) |
## More
You can also download the model weights from following sources:
- BaiduNetdisk: https://pan.baidu.com/s/1gD-NJJWOtaHCtB0qHE79rA with access code **s757**