Our village director’s surname is Li, so we usually call him Director Li. Li’s distant cousin is Sun Jianguo, who lives at the entrance to the village. Sun Jianguo owed money to qian Jianguo of his village. This year thinking of fish to pay off debts. Call Zhou Jianguo, who sells fish fry in the next village. ‘It’s going to take some time to grow into fry,’ Mr. Zhou said. ‘It’s not there now.’ But, but Zhou jianguo sent some photos to Sun Jianguo and told him to sell them like this.
According to the rules, fry are fifty cents a tail. Sun Jianguo thought, I need a large amount, can’t you say that a few tail is a few tail, in case more money to do. Sun Jianguo went to the village director Li. Director Li is also in trouble. Look at the picture. How can I count this? Director Li had a bright brain, and suddenly it occurred to him that SOME time ago I had repaired the camera at the entrance of the village for the village secretary (you can check another blog “The Third Watch, Monocular Camera Calibration Practice (complete process)”). With the mentality of trying, he came to my house to find me.
After seeing the photo, I patted my chest and told Director Li that there was no problem. Now that artificial intelligence is so developed, nothing is impossible. Director Li saw that I was full of confidence and went back full of joy. He told me to wait for his notice before leaving.
Problem analysis
This is an image – based target detection task. Traditional methods and deep learning methods can be used. Traditional methods pose little challenge. But in the spirit of keeping pace with The Times, I prefer the method of deep learning. With deep learning, one of the big challenges is data annotation.
The data indicate
I chose two data annotation tools
(1) Labelimg
Labelimage is relatively easy to use. When you see the fry, click and drag to generate the corresponding target box, and click to save. Not much said. Look at the picture.
(2) Labelme
Why do we use Labelme to label semantically segmented masks? Just trying to think differently.
It’s a bit of a hassle to install. I installed it using Anaconda. For details on how to set up your environment, check out my other blog post deep Learning Walkthroughs! (Series 6: CUDA10.1+ TensorFlow +VS+ Anaconda3 installation).
After the configuration is complete, perform the following operations
1. conda activate tensorflow-gpu
2. conda install pyqt
3. pip install labelme -i https://pypi.tuna.tsinghua.edu.cn/simple
4 not busy exit, in the command good input labelme, you can.
After labeling one fry, I gave up.
No wonder data tagging engineers are paid so much, it makes sense. Life is supposed to be so good, to take on a job like this.
If you have any good tools, you can refer to the public account “Muggle intelligence” reply. Director Li will be here in a few days. Quite urgent.
