DeepImageSearch is a powerful Python library that combines state-of-the-art computer vision models for feature extraction with highly optimized algorithms for indexing and searching. This enables fast and accurate similarity search and clustering of dense vectors, allowing users to build scalable image search systems capable of handling large-scale datasets. The library offers seamless integration with Python and provides GPU support for accelerated processing, delivering a comprehensive solution for researchers and developers working on image-based search and retrieval applications. By incorporating the Vision Transformer (ViT) model, DeepImageSearch further enhances its capabilities in identifying and understanding complex image patterns, making it an essential tool for advanced image search and analysis tasks.
This library is compatible with both windows and Linux system you can just use PIP command to install this library on your system:
pip install DeepImageSearch --upgradeIf you're using a GPU, first uninstall the faiss_cpu version and then try installing the faiss_gpu version. The library installs the CPU version by default because not all systems support GPUs.
How To Use?We have provided the Demo folder under the GitHub repository, you can find the example in both .py and .ipynb file. Following are the ideal flow of the code:
from DeepImageSearch import Load_Data, Search_Setup# Load images from a folderimage_list = Load_Data().from_folder(['folder_path']) # Set up the search engine, You can load 'vit_base_patch16_224_in21k', 'resnet50' etc more then 500+ modelsst = Search_Setup(image_list=image_list, model_name='vgg19', pretrained=True, image_count=100)# Index the imagesst.run_index()# Get metadatametadata = st.get_image_metadata_file()# Add new images to the indexst.add_images_to_index(['image_path_1', 'image_path_2'])# Get similar imagesst.get_similar_images(image_path='image_path', number_of_images=10)# Plot similar imagesst.plot_similar_images(image_path='image_path', number_of_images=9)# Update metadatametadata = st.get_image_metadata_file()This code demonstrates how to load images, set up the search engine, index the images, add new images to the index, and retrieve similar images.
Note: Some models may not work properly due to resizing and normalization issues. By default, I have chosen a size of 224x244. Please try to select models that support this size or resized inputs. I have already tested many models, but testing over 500 is beyond my scope.
DocumentationThis project aims to provide a powerful image search engine using deep learning techniques. To get started, please follow the link: Read Full Documents
Screenshot
If you use DeepImageSerach in your Research/Product, please cite the following GitHub Repository:
@misc{TechyNilesh/DeepImageSearch, author = {VERMA, NILESH}, title = {Deep Image Search - AI-Based Image Search Engine}, year = {2021}, publisher = {GitHub}, journal = {GitHub repository}, howpublished = {\url{https://github.com/TechyNilesh/DeepImageSearch}},}Please do STAR the repository, if it helped you in anyway.More cool features will be added in future. Feel free to give suggestions, report bugs and contribute.