DESCRIPTION: This project is using pretrained model at: https://github.com/filipradenovic/cnnimageretrieval-pytorch on Oxford5k dataset - A dataset of building, so that it would only perform accurately with building images when testing.
In order to run this app you will need
- Docker + WSL2
- pip
- All the rest (data + networks) is automatically downloaded with our scripts
- Step 0: Download and install VSCode, install "Live server" extension to host the UI, start docker desktop
- Step 1: Clone the repo
- Step 2: Run
cd CS419to access into the repo's directory - Step 3: Run
source start_app.shto start the app image and automatically download the dataset to test (default is oxford5k dataset) - Step 4: Open file index.html with Live server extension to start testing
- Step 5: Choose image, crop the interested area and hit the submit, then wait about a second to see the retrieved result
The default number of image to be returned is 16, so the backend will automatically return top 16 matched images. The mechanism is letting backend to extract feature vector for each of images in the dataset (took long time), then save these features in backend/features.pt along with all_images.txt is the image file paths respectively. The current feature.pt & all_images.txt are belonging to oxford5k dataset, if you would like to using another dataset, please follow the instrucion to prepare the features:
- Step 1: Place dataset folder which contains images into backend/data
- Step 2: Adjust the DATASET_PATH to recently added image folder
- Step 3: Run
cd backend; docker-compose -f docker/docker-compose.yml up --buildto rebuild and start backend - Step 4: Run
curl --location --request POST 'localhost:8000/app/v1/extract-feature'to start extracting dataset - Step 5: If extract process successfully, you can start tesing new dataset and images
Cong Minh Tran - congminht91@gmail.com | minhtc6@viettel.com.vn
Project link: https://github.com/minhct13/CS419
=======
38272e953434cf97d04e7f5fa3934c7ce9df5cf0
DESCRIPTION: This project is using pretrained model at: https://github.com/filipradenovic/cnnimageretrieval-pytorch on Oxford5k dataset - A dataset of building, so that it would only perform accurately with building images when testing.
In order to run this app you will need
- Docker + WSL2
- pip
- All the rest (data + networks) is automatically downloaded with our scripts
- Step 0: Download and install VSCode, install "Live server" extension to host the UI, start docker desktop
- Step 1: Clone the repo
- Step 2: Run
cd CS419to access into the repo's directory - Step 3: Run
source start_app.shto start the app image and automatically download the dataset to test (default is oxford5k dataset) - Step 4: Open file index.html with Live server extension to start testing
- Step 5: Choose image, crop the interested area and hit the submit, then wait about a second to see the retrieved result
The default number of image to be returned is 16, so the backend will automatically return top 16 matched images. The mechanism is letting backend to extract feature vector for each of images in the dataset (took long time), then save these features in backend/features.pt along with all_images.txt is the image file paths respectively. The current feature.pt & all_images.txt are belonging to oxford5k dataset, if you would like to using another dataset, please follow the instrucion to prepare the features:
- Step 1: Place dataset folder which contains images into backend/data
- Step 2: Adjust the DATASET_PATH to recently added image folder
- Step 3: Run
cd backend; docker-compose -f docker/docker-compose.yml up --buildto rebuild and start backend - Step 4: Run
curl --location --request POST 'localhost:8000/app/v1/extract-feature'to start extracting dataset - Step 5: If extract process successfully, you can start tesing new dataset and images
Cong Minh Tran - congminht91@gmail.com | minhtc6@viettel.com.vn
Project link: https://github.com/minhct13/CS419
38272e953434cf97d04e7f5fa3934c7ce9df5cf0