This is the repository for the Multimedia Information Retrieval Course Final Project Mineral Fashion Image Retrieval System. This project was carried out by a group of students from the Falcuty of Computer Science at University of Information Technology.
This system was trained on the FashionIQ dataset, showing its applicability to the fashion domain for conditioned retrieval, and to more generic content considering the more general task of composed image retrieval.
To install and use this system demo please follow these simple steps.
- Clone the repo
git clone https://github.com/emerald-lan/CS336.O11.KHCL.git- Install dependencies
pip install -r requirements.txt- Data Preparation
The dataset and necessary files will be downloaded and prepare automatically by using the following command:
python src/download_resources.pyAfter the running of src/download_resources.py is done, the splited_fashionIQ folder (containings images for retrieval purposes), finetuned_RN50.pt and filtered_gallery.json file will be downloaded and extracted. Everything should be up-to-date now.
Here's a brief description of every important file/folder:
data: Folder contains images, indices and captions.src: Folder contains important resources.src/utils: Folder contains searching and feature extracting modules.src/resources: Folder contains fine-tuned file.src/download_resources.py: Download necessary resourses.src/dataset_process.py: Process dataset for purposes.src/dataset_process.py: Process dataset for purposes.
test: Folder contains python notebooks for testing purposes.test/test_reformulate: Reformulating the retrieval results using users' feedbacks (underdeveloped).
app.py: Running the application.requirements.txt: Contains dependencies
Start the server and run the demo using the following command
python app.pyBy default, the server run on port 8501 of localhost address: http://localhost:8501/