README.md

## Introduction

<a href="https://github.com/facebookresearch/MaskFormer">Official Repo</a>

<a href="https://github.com/SegmentationBLWX/sssegmentation/blob/main/ssseg/modules/models/segmentors/maskformer/maskformer.py">Code Snippet</a>

<details>
<summary align="left"><a href="https://arxiv.org/pdf/2107.06278.pdf">MaskFormer (NeurIPS'2021)</a></summary>

```latex
@inproceedings{cheng2021per,
    title={Per-pixel classification is not all you need for semantic segmentation},
    author={Cheng, Bowen and Schwing, Alex and Kirillov, Alexander},
    booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
    year={2021}
}
```

</details>


## Results

#### ADE20k

| Segmentor     | Pretrain               | Backbone    | Crop Size  | Schedule                                | Train/Eval Set  | mIoU   | Download                                                                                                                                                                                                                                                                                                                                                                                                |
| :-:           | :-:                    | :-:         | :-:        | :-:                                     | :-:             | :-:    | :-:                                                                                                                                                                                                                                                                                                                                                                                                     |
| MaskFormer    | ImageNet-1k-224x224    | Swin-Tiny   | 512x512    | LR/POLICY/BS/EPOCH: 0.00006/poly/16/130 | train/val       | 47.31% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/maskformer/maskformer_swintiny_ade20k.py) &#124; [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_maskformer/maskformer_swintiny_ade20k.pth) &#124; [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_maskformer/maskformer_swintiny_ade20k.log)    |
| MaskFormer    | ImageNet-1k-224x224    | Swin-Small  | 512x512    | LR/POLICY/BS/EPOCH: 0.00006/poly/16/130 | train/val       | 49.91% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/maskformer/maskformer_swinsmall_ade20k.py) &#124; [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_maskformer/maskformer_swinsmall_ade20k.pth) &#124; [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_maskformer/maskformer_swinsmall_ade20k.log) |
| MaskFormer    | ImageNet-22k-384x384   | Swin-Base   | 640x640    | LR/POLICY/BS/EPOCH: 0.00006/poly/16/130 | train/val       | 53.22% | [cfg](https://raw.githubusercontent.com/SegmentationBLWX/sssegmentation/main/ssseg/configs/maskformer/maskformer_swinbase_ade20k.py) &#124; [model](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_maskformer/maskformer_swinbase_ade20k.pth) &#124; [log](https://github.com/SegmentationBLWX/modelstore/releases/download/ssseg_maskformer/maskformer_swinbase_ade20k.log)    |


## More

You can also download the model weights from following sources:

- BaiduNetdisk: https://pan.baidu.com/s/1gD-NJJWOtaHCtB0qHE79rA with access code **s757**