Customized implementation of R-CNN (Regions with Convolutional Neural Network Features) algorithms.
Find Out MoreDeep Learning Neural Networks have been commonly used in the field of object recognition. R-CNN has successfully applied convolutional neural network to build a strong and robust algorithms in 2014. In this project the ultimate objectives are to: 1) reproduce R-CNN on Python; and 2) replace original SVM classifier with LDA classifier to improve on accuracy rate. To achieve the goal, project team will utilize ImageNet ILSVC 2012 and PASCAL VOC 2007 datasets to train the model and leverage PASCAL VOC 2012 dataset to evaluate the performance.
Learn more about R-CNN
RCNN is using a modular disign, so we reimplement the algorithm by modules and conduct testing seperately.
We have studied some open source codes when implementing selective search and make our own version based on the studies.
We use Caffe to implement the CNN module. Because the original RCNN is also using Caffe, we make use of their Caffe configuration and change a little bit.
● Final report submission
● Final project presentation
● Final presentation poster design
● Final project exhibition
Workable Python implementation of R-CNN. All modules have been tested and evaluated seperately.
Toy program practicing Topic model(LDA).
