Last page: wiki: Strat pytorch
Two easy ways to install zisan. Input the follow command:
pip install zisan
Also you can download the wheel file on the official site: http://jintupersonal.com/zisan/
Pay attention to the lastest version to modify your pip command.
Firstly you should download the weights files for zisan. You can find the url button on homepage: Download weights link:https://pan.baidu.com/s/1qj-Lpe4OKV0L-w9uKO8EFw key:x9wl The weight files structure is followed:
Instance_Seg_weight (Folder)
Jintu_SEG_Interactive.pth (178 MB)
ObjectDetect_data_weights (Folder)
runBox.zip (475 MB)
Unzip runBox.zip, you will see three folders, they are "cfgs", "data", "weights".
cfgs: Here put the network configs
data: Put your data preparing to train
weights: Here save the raw yolo weights and your last output weights
DON'T RECHANGE THE FOLDER'S NAME !
DON'T RECHANGE THE FOLDER'S NAME !
DON'T RECHANGE THE FOLDER'S NAME !
you can't rechange the "cfgs", "weights", "data" and their child folders name, because zisan. Objedetect class will use the refer all folders' name, if you rechange them, it will happen errors. But you can set your current path and as parameters in function.
more details you can refer the following page: wiki: Train your dataset
How to use these weights? You will find your answer in the following Demo Courses.
Next page: wiki: Class ImgSeg
Raw page: wiki: Demo: A person segmentation Last page: wiki: Demo: A Box segmentation
(This picture is from dataset davis2017) Find the person and give him bone marks, like the following. Maybe it's very abstract. Monofilament doesn't affect our segmentation of objects。
from zisan.Seg.Interface import ImgSeg, markTools
import os
import numpy as np
import cv2
from PIL import Image
lines=[[(281,120),(267,341)],[(279,157),(208,171)],[(309,170),(308,250)],[(275,233),(370,341)]]
img=Image.open(current_path+'/temp/1.jpg').convert('RGB')
markpen=markTools(img.height,img.width)
for line in lines:
markpen.curveDraw(line,is_Pos=True)
[(281,120),(267,341)] is a line element lines is a line list=[[(,),(,)]]
re=markpen.getMark_result(is_showPreview=True)
model.ImgSeg_SingleObj(img,re,is_showPreview=True)
Fade mask show preview:
zisan have the Yolov3 interface to train your own dataset. Please view the next part: wiki: Package: Object detect