How xmu-xiaoma666/external-attention-pytorch works
FightingCV is a PyTorch component library for reusable computer-vision building blocks, not an end-to-end training application. Its documented user workflow is to import a component, construct it with task-specific dimensions, pass tensors through it, and receive transformed tensors. The README groups the library into attention, backbone, MLP, convolution, and re-parameterization families; these nodes represent those substantial capability areas. Source evidence is limited, so internal links between library families are omitted rather than inferred.
Main entry points: MobileViTv2Attention.py MobileViTv2 attention · main.py Example runner · attention Attention components · backbone Vision backbones
Components 9
Attention
- Attention components
model/attention - MobileViTv2 attention
model/attention/MobileViTv2Attention.py
Backbones
- Vision backbones
model/backbone
MLP Blocks
- MLP components
model/mlp
Convolution
- Convolution components
model/conv
Re-parameterization
- Re-parameterized blocks
model/rep
Other components
- Library user
- Example runner
main.py - Output tensor
Connections 7
- Library user → Attention components: imports and uses
README.md - Library user → Vision backbones: uses models
README.md - Library user → MLP components: uses blocks
README.md - Library user → Convolution components: uses modules
README.md - Library user → Re-parameterized blocks: uses blocks
README.md - Example runner → MobileViTv2 attention: instantiates and calls
main.py - MobileViTv2 attention → Output tensor: returns
model/attention/MobileViTv2Attention.py