◐ModelVision & ImageFree
MMSegmentation
OpenMMLab's unified semantic segmentation toolbox with 40+ architectures (DeepLab, SegFormer, Mask2Former) and 250+ pretrained models.
インストール数128k
評価★ 4.6
レビュー43
MMSegmentation
MMSegmentation is OpenMMLab's open-source semantic segmentation toolbox built on PyTorch. It implements a clean abstraction over backbone, neck, decode head, and loss components, making it trivial to combine architectures or benchmark new ideas against state-of-the-art baselines.
Key Features
- 40+ architectures — FCN, PSPNet, DeepLabV3+, EncNet, PointRend, SegFormer, Mask2Former, ISNet, and more
- 250+ pretrained models — weights from ADE20K, Cityscapes, VOC, LoveDA, iSAID, and other benchmarks
- Modular design — swap backbone (ResNet, Swin, ViT, ConvNeXt), neck, and head independently via config YAML
- Multi-scale testing — built-in flip, scale, and sliding-window inference modes
- Mixed precision — automatic FP16 training with
--amp - Distributed training — SLURM and
torchrun-based multi-GPU/multi-node support - mmseg-infer — one-line CLI for inference on images, folders, or video
Quick Start
pip install -U openmim && mim install mmengine mmsegmentation
mmseg-infer demo/demo.png configs/segformer/segformer_mit-b0_8xb2-160k_ade20k-512x512.py \
--weights https://download.openmmlab.com/mmsegmentation/v0.5/segformer/... \
--out-dir outputs/
Install via ai-supply
npx ai-supply add mmsegmentation-semantic-segmentation
Curated mirror of the open-source MMSegmentation (Apache-2.0). Get it from the source.