yolov2

A simple implementation of yolov2 based on tensorflow Only for detection, training is not supported The repo is based on https://github.com/KOD-Chen/YOLOv2-Tensorflow, I just use the weights from official yolo page and for private usage

HOW TO USE

dev:
compressor.py: testing different compression methods for feature maps
gen_results.py: generate the results for testing AP by using those feature maps compression methods
plot.py: plot figures

client:
preprocess.py: run the first part of yolov2 generate byte stream and output to UDS socket

model:
model files

1. git clone https://github.com/thtrieu/darkflow/tree/master/darkflow to install darkflow(follow the official guide)
2. download weights from https://pjreddie.com/media/files/yolov2.weights and download cfg file from https://github.com/pjreddie/darknet/blob/master/cfg/yolov2.cfg
3. run flow --model cfg/yolo.cfg --load bin/yolo.weights --savepb by which --model and --load specify the cfg and weights files downloaded
4. under built_graph dictory will be the model files and just put them under the model dictory of this project

theRest:
testing some unusual methods for feature maps compression like using autoencoders and compressed sensing

tools:
scripts for memory and cpu usage profiling and plot setup definition(tex.py)

Run in docker

make sure docker is installed, then

1. cd dockerfiles && docker build -t yolov2 -f Dockerfile.yolov2 .

2. 5 minutes waiting and everything is there

3. docker run -it ID /bin/bash