A compact Convolutional Neural Networks based model will be developed to detect known objects in an image in real time taking inspiration from SqueezeNet, SqueezeDet and using tiny-dnn deep learning library. Regression based model of YOLO (You Only Look Once) is incorporated to localize for objects in the image and generate bounding boxes. Fully connected layers at the end of the pipeline of YOLO are replaced with a single convolutional layer "ConvDet layer" for generating bounding boxes. Further Deep Compression will be implemented which includes methods like Network Pruning, Quantization and Huffman Coding to reduce the size of the model even further while still maintaining the baseline performance.

Organization

Student

kv

Mentors

  • Vladimir Tyan
close

2017