watersink / deeplearningdemo

CV deeplearning demo

what is it

Collect all kinds of deep learning projects, quickly build demo. Use the minimum code to Reproducing function

It can be divided into CPU version and GPU version. Some project will adopt GPU version which the model can only run on GPU or the model with poor experience on CPU

• Small model use cpu inference

• Large model uses docker container inference,and destroy the container after completion

[中文说明] [deploy] [video tutorial]

sample

openface

detecron2

face 3D reconsturction

demo list

name	inference	?pu	model	source
face detection	memory	cpu	mtcnn	-
key point	container	cpu	openface	github
face alignment	container	cpu	face-alignment	-
segmentation	container	gpu	detecron2	github
segmentation	container	cpu	maskrcnn	github
pose estimation	container	gpu	densepose	github
makeup	container	gpu	face-makeup.PyTorch	github
attribute predict	memory	cpu	attribute-predict	-
face verification	memory	cpu	arcface	github
face 3D reconsturction	container	cpu	prnet	github
super resolution	container	cpu	srgan	github
AImakeup	memory	cpu	beautygan	github
attribute manipulate	container	gpu	stylegan	github
attribute manipulate	container	gpu	glow	github
face interpolation	container	gpu	glow	github
face interpolation	container	gpu	stylegan	github
image inpaint	-	web	-	nvidia-playground
domain transfer	-	web	guagan	nvidia-playground
domain transfer	-	web	gannimal	nvidia-playground
tf-js	-	web	-	github

Install

• main image

docker pull zzz9958123/demo_server

• project image

docker pull zzz9958123/glow
docker pull zzz9958123/detectron2
docker pull zzz9958123/densepose
docker pull zzz9958123/openface
docker pull zzz9958123/densepose
docker pull zzz9958123/maskrcnn-benchmark
docker pull zzz9958123/prnet
docker pull zzz9958123/haircolour
docker pull zzz9958123/face_alignment
docker pull zzz9958123/srgan

• download model from [GD] [baidu]

• Place your domain certificate in the SSL folder to make HTTPS work.This is not a necessary step, unless you want the webcam to work

Run

docker-compose up

go check http(s)://0.0.0.0:5000

rest-api

All projects support rest API methods
URL is the project address,Post requests to submit form, input {Base64: Base64 picture}, and return JSON calculation result. Refer to response JSON of demo page for structure

example

import requests
import base64
res = requests.post("https://pc.zzz9958123.com:5000/attribute_predict",data={"base64":b"data:img/png;base64,"+base64.b64encode(open("abc.png","rb").read())}).content
print(res)