metafile.yml

Collections:
  - Name: MaskFormer
    Metadata:
      Training Data: COCO
      Training Techniques:
        - AdamW
        - Weight Decay
      Training Resources: 16x V100 GPUs
      Architecture:
        - MaskFormer
    Paper:
      URL: https://arxiv.org/pdf/2107.06278
      Title: 'Per-Pixel Classification is Not All You Need for Semantic Segmentation'
    README: configs/maskformer/README.md
    Code:
      URL: https://github.com/open-mmlab/mmdetection/blob/v2.22.0/mmdet/models/detectors/maskformer.py#L7
      Version: v2.22.0

Models:
  - Name: maskformer_r50_ms-16xb1-75e_coco
    In Collection: MaskFormer
    Config: configs/maskformer/maskformer_r50_ms-16xb1-75e_coco.py
    Metadata:
      Training Memory (GB): 16.2
      Epochs: 75
    Results:
    - Task: Panoptic Segmentation
      Dataset: COCO
      Metrics:
        PQ: 46.9
    Weights: https://download.openmmlab.com/mmdetection/v3.0/maskformer/maskformer_r50_ms-16xb1-75e_coco/maskformer_r50_ms-16xb1-75e_coco_20230116_095226-baacd858.pth
  - Name: maskformer_swin-l-p4-w12_64xb1-ms-300e_coco
    In Collection: MaskFormer
    Config: configs/maskformer/maskformer_swin-l-p4-w12_64xb1-ms-300e_coco.py
    Metadata:
      Training Memory (GB): 27.2
      Epochs: 300
    Results:
    - Task: Panoptic Segmentation
      Dataset: COCO
      Metrics:
        PQ: 53.2
    Weights: https://download.openmmlab.com/mmdetection/v3.0/maskformer/maskformer_swin-l-p4-w12_64xb1-ms-300e_coco/maskformer_swin-l-p4-w12_64xb1-ms-300e_coco_20220326_221612-c63ab967.pth