imSearch: A generic framework to build your own reverse image search engine

imsearch helps to create your own custom, robust & scalable reverse image search engine. This project uses state of the art object detection algorithm (yolov3) at its core to extract the features from an image. It uses an efficient cross-platform similarity search library NMSLIB for similarity search. Redis is used as a messaging queue between feature extractor and core engine. MongoDB is used to store the meta-data of all the indexed images. HD5 file system is used to store the feature vectors extracted from indexed images.

Installation

For the setup, a simple install.sh script can be used or can be installed using pip. Follow these simple steps to install imsearch library.

• Feature extraction is GPU intensive process. So, to make the search real-time, running this engine on GPU enabled machine is recommended.
• Install CUDA & NVIDIA graphics drivers (here)
• Install PyTorch (here)
• Install MongoDB (here)
• Install Redis (here)
• Run following commands

pip install --no-binary :all: nmslib
pip install imsearch

Build from source using install.sh

git clone https://github.com/rikenmehta03/imsearch.git
chmod +x install.sh
./install.sh

Example usage

import glob
import imsearch

all_images = glob.glob('path/to/image/folder')

# Initialize the index
index = imsearch.init('test')

# Add single image to the index
index.addImage(all_images[0]) 

# Add multiple image to the index
index.addImageBatch(all_images[1:])

# Create index and make it ready for the search query
index.createIndex() 

# find k nearest similar images
# choose policy from 'object' or 'global'. Search results will change accordingly.
# object: Object level matching. The engine will look for similarity at object level for every object detected in the image.
# global: Overall similarity using single feature space on the whole image. 
similar = index.knnQuery('path/to/query/image', k=10, policy='object')

For detailed usage see examples/index.py

Credit

YOLOv3: An Incremental Improvement

Joseph Redmon, Ali Farhadi

Abstract
We present some updates to YOLO! We made a bunch of little design changes to make it better. We also trained this new network that’s pretty swell. It’s a little bigger than last time but more accurate. It’s still fast though, don’t worry. At 320 × 320 YOLOv3 runs in 22 ms at 28.2 mAP, as accurate as SSD but three times faster. When we look at the old .5 IOU mAP detection metric YOLOv3 is quite good. It achieves 57.9 AP50 in 51 ms on a Titan X, compared to 57.5 AP50 in 198 ms by RetinaNet, similar performance but 3.8× faster. As always, all the code is online at https://pjreddie.com/yolo/.

[Paper] [Project Webpage] [Authors' Implementation]

@article{yolov3,
  title={YOLOv3: An Incremental Improvement},
  author={Redmon, Joseph and Farhadi, Ali},
  journal = {arXiv},
  year={2018}
}

PyTorch-YOLOv3

Minimal PyTorch implementation of YOLOv3 [GitHub]