This repo is focus on yolov3-tiny

After enable fixed point calculation in Makefile (fixed point only support on CPU, pls set GPU=0 )
GEMM_FIXED=1, BIAS_FIXED=1

fixed point calcuation is simulate by int64 & convert back to floating point after each convolution

explain how to build up dataset

As here https://pjreddie.com/darknet/yolo/

As here https://pjreddie.com/darknet/yolo/

convert by custem code
updating... 
[Todo] collecting convertion code & make doc.

download at: https://www.zybuluo.com/nrailgun/note/488084 <br>
follow by : https://pjreddie.com/darknet/imagenet/ <br>

Default with yolov3-tiny or yolov3 <br>
you can get more yolo model at https://pjreddie.com/darknet/yolo/ <br>
get model for image classify at https://pjreddie.com/darknet/imagenet <br>

how to train your own model...   <br>
[Todo] make doc.

We need to decide fractional point for each layer by some data.
Set GEMM_FIXED & BIAS_FIXED =1 in Makefile
Set find_new_fixed=1 in index.sh and index xx -b
Run ./index.sh kitti_small -m #run 20 image for decide fractional point

Set find_new_fixed=0 in index.sh and index xx -b
Run project with fixed point config file.

Run yolov3-tiny for different config
./index.sh [kitti/voc/coco] -[?/i/m/t/...]

Choose the dataset you want and the opration to do.
You need to build up dataset first.

image classify
./darknet classifier predict cfg/imagenet1k.data cfg/darknet19.cfg darknet19.weights ./darknet classifier valid cfg/imagenet1k.data cfg/darknet19.cfg darknet19.weights

more function is comming soon... [Todo] fixed point policy [Todo] modify policy for input_data_flow