Here are my projects. Some of them are my academy research, and the others are just my interest.
My role: Full developer
VeReS is an vehicle retrieval system particularly for Vehicle search. The system can be used to find the same car or car re-identification. The feature model is learned by metric learning, and detection is applied before feature is extracted in the boundingbox, Developing the system is just for interest in my free time. Opinions are solely my own.
Technologies: Metric Learning, TensorFlow, C++, QT, OpenCV.
My role: Full developer
INS is an image retrieval system for instance search. The system can be used to retrieve same object, near-duplicate object, and copy detection, and developing the system is just for interest in my free time. I'll improve the system contiously.
Technologies: CNN, C++, QT, OpenCV, Query Expansion
My role: Main developer
Project: cnn-cbir-benchmark
CNN CBIR Benchmark tries to build a benchmark for image retrieval, particularly for object retrieval. The project has finished three CNN-based CBIR methods and one hand-crafted feature (i.e. fisher vector) method. The project has just released one FC image retrieval method. Everything is still going on.
Technologies: CNN features, Hand-crafted features
My role: Full developer
Code: SeetaFaceLib
SeetaFaceLib is face image retrieval and recognition project based on SeetaFaceEngine. I work in a Mac OS envirment, so I'd like to make the SeetaFaceEngine work in Mac OS. I have make some changes to make it work in Xcode. More importantly, I have also built a QT project of face image retrieval.
Technologies: Image Retrieval, Face Recognition, OpenCV, C++, QT
My role: Main developer
The information of pupilTracker will be coming soon.
Technologies: Image Proccessing, OpenCV, C++
My role: Develop independently
Convolutional Neural Network has been sucessfully applied to image classification, object detection, etc. However, It's not easy to apply it to duplicate search or object retrieval due to the difficult of data enhancement. Duplicate search can be used to many business applications such as searching the same clothes or shoes on Taobao, and I believe it has great application vulae. So after diving into image retrieval based on key point feature such as SIFT, I implement a prototype system based on bag of feature. The performance of the prototype image retrieval system works very well, and the mAP can be achieve to 83.35%. Moreover. I test it on 1500,000 clothes images, and the search result can be found here.
Technologies: SIFT, BoVW, Image Retrieval
Video: natural images, medical images
Code: feature extraction, search online
My role: Develop independently
After researching CBIR one year, I determined to develop a image retrieval engine demo just for interest. The convolutional neural networks in the comuter vision community develop so fast in recent years and show great performance with various tasks. So in my spare time I dive to read papers of CNN and program. The dataset of PicSearch engine is Caltech256, which contains 29780 images, and until now it supports query in the dataset only.
Technologies: CNN, Python, Image Retrieval
Blog post: Image retrieval using MatconvNet and pre-trained imageNet (Chinese)
My role: Develop independently with open sources
I do the image retrieval research in early 2013. My mission is to design efficient hashing algorithm to map the semantic similar images to the similar codes. There are two main advantages using hashing method for image retrieval, i.e. storage and computation efficience. To let more researchers focus on design hashing algorithm, I have built a hashing baseline, hoping this project can do some help for some researchers.
Technologies: Matlab, Hashing, Image Retrieval
My role: equal contribution with one partner
I'm a big fun of Python. I think Python is one of the most elegant programming language
I have known. It's a great power tool which can be used everywhere, such as write scripts,
using Django for web developmentand do science computation. When I first meet the book
Technologies: Python, OpenCV.