This repo provides support for using NetVLAD for inference only purpose with a single line of code (aka torch hub).
pip install -r requirements.txt
model = torch.hub.load("jarvisyjw/netvlad", 'my_model', model_name=VGG16-NetVLAD-Pitts30K, whiten=True)
- A retrieval example is provided in
test_model.py
python test_model.py
Gives results: