Netvlad using lightweight framework mobilenet_v2 for feature extraction！The repository includes :
- Inference demo
- Pre-trained model
- Sample image
- Totorial and description
The repository is created by Ze Huang
First, make sure you have installed python 3.8 environment and downloaded this project repository.
Next, run the following command in the home directory to install the python package needed to use this model
pip install -r requirements.txt
Run python demo.py to get the feature and vlad vector of the sample image without CUDA.
You can also specify the image path or using CUDA by runing the following command.
python demo.py --image='imagepath' --cuda
| item | value |
|---|---|
| model | mobilenet_v2 + netvlad |
| size | 18.50M |
| number of clusters | 64 |
| training data | Pittsburgh |
| accuracy | R@1=80.07% R@5=91.25% R@10=93.99% |
| time of load network | cpu=0.111s gpu=0.148s |
| time of inference(single image) | cpu=0.136s gpu=0.020s |
| GPU | TITAN XP |