PinkJelly

Instant-object-recognition && speech App, easy as 1, 2 ,3!

Quick Start

git clone pinkJelly

https://github.com/frozenabe/PinkJelly.git

Inside server folder, git clone Darknet yolo2

website : https://pjreddie.com/darknet/yolo/
git : git clone https://github.com/pjreddie/darknet

Getting weights inside Darknet folder

yolo2 : wget https://pjreddie.com/media/files/yolo.weights
tiny-yolo : wget https://pjreddie.com/media/files/tiny-yolo-voc.weights
yolo9000 : wget https://pjreddie.com/media/files/yolo9000.weights

Replace image.c with darknet/src/image.c

Download our App using Expo

https://exp.host/@zao1214/client ( if for some reason doesn't work, then it is probably b/c we stopped the server. Give us a notice and we will be happy to re-run the server! )

then follow the steps

1. login (Email confirmation required) for signup
2. take a picuture (simple version comes with 200 objects quite accurate, funny comes with 9400 but less accurate)
3. experience the magic (press the circular-object-detected-button and hear its label in English)

!IMPORTANT notice :

• currently, machine learning behind simple camera is tiny-yolo (current server with 1G memory can only hold tiny, which also means it will crash on funny(yolo9000) camera)
• In order to change yolo version, code can be found in runDarknet.js where I have written some comment
• If we replace AWS EC2 server to c4.large, yolo2 and yolo9000 can be used and it will take 8-10 seconds
• If we replace AWS EC2 server to any GPU server, it takes 3-4 seconds
• If we change server code to preload weights on server, every process will take 3seconds less, (GPU 0-1seconds) (I personally like to have loading section, so I decided not to put weights on)

Credit :
Darknet@article{redmon2016yolo9000,
  title={YOLO9000: Better, Faster, Stronger},
  author={Redmon, Joseph and Farhadi, Ali},
  journal={arXiv preprint arXiv:1612.08242},
  year={2016}
}