While go through vacation photos, people always ask themself: What is the name of this temple I visited in Thailand? Who created this monument in France? Landmark recognition can help! This technology can predict landmark labels directly from image pixels, to help people better understand and organize their photo collections.
In this project, a model use for landmark recognition will be built based on ImageNet Classification With Deep Convolutional Neural Networks
- Python 3.x
- Install the requirements packages
pip install -r requirements.txt
- The dataset can be downloaded from Google Landmark Recognition Challenge.
- Unzip the downloaded dataset and put under
src/data/dataset.csv
and follow the notebooksrc/analysis.ipynb
or just use the reduced version of the original dataset. - Use the script
src/utils/download_data.py
to download the images. Use the command line way:python src/utils/download_data.py "src/data/train.csv" "src/data/train"
or just follow the notebooksrc/analysis.ipynb
src/data/train.csv
- the training images setsrc/data/valid.csv
- the validation images setsrc/data/test.csv
- the test images setsrc/analysis.ipynb
- The notebook that I've used to analyze and preprocess datasetsrc/benchmark_model.ipynb
- The notebook that I've used to build a benchmark modelsrc/vgg16_model.ipynb
- The notebook that I've used to build my final solution