Image Retrieval

This is a project for image retrieval implemented in C++/Python. It consists of two main parts: feature extraction and vector retrieval.

Prerequisites

C++

Python

Getting Started

ImageNet-1K

Download ImageNet-1K from Hugging Face

Feature Extraction

Extract image features using pre-trained ResNet50.

python image_retrieval/feature_extraction/inferencer.py \
    -d /path/to/imagenet_1k_rawimgs \
    -i /path/to/imagenet_1k_rawimgs/train.txt \
    -o data.pb \
    -j image_retrieval/feature_extraction/resnet50.py \
    -w 8 \
    -b 64

Vector Search

Once the features are extracted, they are transformed into a vector representation. This allows for efficient indexing and searching of images based on their visual similarities.

./image_retrieval/ann/search_engine -i data.pb -p 8001

Demo UI

python image_retrieval/demo.py 8000 -t localhost:8001 --resource /path/to/imagenet_1k_rawimgs