• Google Landmark Recognition
• • Acknowledge
  • Environment:
  • Train Data preparation
  • Training
  • Test Data preparation
  • Predicting

Google Landmark Recognition

Acknowledge

This work was supported by the National Key R&D Program of China under Grant No. 2020AAA0103804(Sponsor: Hefu Liu). This work belongs to the University of science and technology of China.

Environment:

1. pip install -r requirements.txt

2. git clone https://github.com/NVIDIA/apex.git & cd apex

3. python3 setup.py install

4. pip install -v --no-cache-dir ./

Train Data preparation

1. Download the Google Landmarks Dataset v2 to ./data using the scripts at https://github.com/cvdfoundation/google-landmark This is our training data.

2. Download the train label csv file from here and put it in ./train.csv

3. Download ReXNet_V1-2.0x pretrained model weights from here and put it in ./rexnetv1_2.0x.pth

4. python preprocess.py to get ./train_0.csv and ./idx2landmark_id.pkl

Training

After training, models will be saved in ./weights/ Tranning logs will be saved in ./logs/ by default.

bash run.sh

Test Data preparation

1. Download test data set, put it in ./data/test, download test.tar.part1, test.tar.part2, then cat test.tar.* | tar -xv

2. python preprocess_test.py to generate ./test.csv label for test, or download from here

3. Download final checkpoint ./weights/final_model.pth from here

Predicting

bash run_test.sh