A variety of free and open source artificial intelligence projects, such as: training a model to let artificial intelligence play King of Glory.
Full text outline
PULSE - This open source project can achieve demosaicing or high-definition by adding pixels to pictures.
Depix - Decodes mosaiced text.
TecoGAN - Demosaicing or super-resolution of videos.
SkinDeep - black technology to remove tattoos from pictures and videos with one click.
StyleCLIP - AI automatic P map, have you seen Musk with the broom head?
polyglot-png - Obviously the download is a picture, just modify the suffix name, the picture becomes a song, a string of codes.
ResnetGPT - Teach you to train a model to let artificial intelligence play King of Glory.
intelligent-uavpath-planning - This project is an intelligent UAV path planning simulation system.
EssayKiller_V2 - One person used three months to develop an artificial intelligence that can write essays.
style2paints - No need to install any environment, do any configuration, download and double-click, you can color the line drawing.
PULSE
Github（7k）：https://github.com/adamian98/pulse
PULSE - This open source project can demosaic or high-definition by adding pixels to the picture
Mosaic: a stumbling block to the progress of human civilization, the mortal enemy of nerds, and the nightmare of old drivers. Since the birth of mosaics, the struggle between humans and mosaics has never stopped.
For programmers, if girls ask us the most question: how to install the system? Then the most frequently asked question by boys must be: how to remove the mosaic.
If someone asks if you can remove the mosaic? You can give him an affirmative answer: no, the mosaic is irreversible.
But, but! You can make up your mind, I have seen too much of that thing, what is the place where the code is coded, do you not know what it is? But for a normal person, what I’m talking about here is a normal person. Even if you read countless photos and give you a severely coded photo of a beautiful woman, it is difficult for you to figure out what this beautiful woman looks like. There are two reasons :
-
You haven't seen enough beautiful pictures
-
You did not establish the link between the beauty mosaic and the original picture of the beauty
Some friends said, I don’t eat or drink, I just look at pictures of beauties and the corresponding mosaics. When I have seen enough and give me a coded picture of a beautiful woman, will it immediately pop up in my mind what this beautiful woman looks like?
Wonderful, at first glance, there is really something. Could it be that "no code in mind is HD"?
As shown in the picture:
Depix
Github (23.1k): https://github.com/beurtschipper/Depix
Depix - Decode mosaic text
As shown in the picture:
TecoGAN
Github（5.1k）：https://github.com/thunil/TecoGAN
TecoGAN demosaic or super-resolution video
As shown in the picture:
SkinDeep
Github（38.3k）：https://github.com/vijishmadhavan/SkinDeep
SkinDeep - black technology to remove tattoos from pictures and videos with one click
Unexpectedly, there is such an open source project on GitHub. Vijish Madhavan, an algorithm researcher from India, has open sourced a machine learning tool SkinDeep, which can automatically remove tattoos from stars.
I have used this tool to process some photos of people with heavy tattoos, and the results are not bad. The picture below shows Allen Iverson (American basketball player) before and after tattoo removal.
As shown in the picture:
StyleCLIP
Github（3.2k）：https://github.com/orpatashnik/StyleCLIP
StyleCLIP - AI automatic P map, have you seen Musk with the broom head?
This project just now showed the power of AI, and the effect is comparable to PS. The next project of automatic P-map is really awesome.
Have you ever thought about such a scene, when you say to an original picture: help me P a handsome bangs, pop! A picture with handsome bangs is automatically generated.
As shown in the picture:
polyglot-png
Github（3.2k）：https://github.com/DavidBuchanan314/tweetable-polyglot-png
polyglot-png - Obviously the download is a picture, just modify the suffix name, the picture becomes a song, a string of codes
Foreign hacker David Buchanan took advantage of Twitter's loopholes to transmit an "encrypted" file by disguising a picture, provided that it does not exceed 3MB.
He managed to compress the GitHub source code of this hidden file into an image.
Now you only need to go to his Twitter, download this picture, and change the file extension from .png to .zip, then you can decompress it into Github code.
The principle of hiding compressed packages with pictures is not complicated. The format of png picture files is as follows. After Zlib, there is a slice of additional data in the IDAT block. The stash data goes here.
Image address with compressed package:
https://i.imgur.com/kNhGrN3.png
David Buchanan's Twitter:
https://twitter.com/David3141593/status/1371974874856587268
As shown in the picture:
ResnetGPT
Github（2.2k）：https://github.com/FengQuanLi/ResnetGPT
ResnetGPT - Teach you to train a model to let artificial intelligence play King of Glory.
What is the implementation principle of this open source project? The figure below is the core code of this model, it is not difficult, I will dismantle it step by step, and I will try my best to describe these technologies in plain language.
Letting computers play games for us can be divided into the following steps. Of course, this method is relatively easy to implement and does not use "reinforcement learning". The AlphaGo algorithm that slaughtered the Go circle is based on reinforcement learning.
-
Get the current game interface. What is the state of the hero now, whether there are enemy heroes and soldiers around, and so on.
-
Generate operation instructions based on the current game interface state, whether to move forward, attack or release skills.
-
According to the generated operation instructions to control the mobile phone, the hero makes corresponding actions.
For training a civilian version of the artificial intelligence model, the general process is like this. Among them, there are mature technologies for obtaining the current game interface and controlling the mobile phone according to instructions.
For example, in this project, scrcpy is used to obtain the projection screen of the Android phone. This Android projection artifact can project the game screen to the computer desktop.
With the game screen, you can get the situation of the battle at this time. The game screen is in the form of a picture. As for what the picture contains, humans can understand it, but computers don't.
This requires us to extract the features in the picture by ourselves, and here we will use the deep learning algorithm. Based on the convolutional neural network (CNN) to extract the features of the picture, the ResNet 101 classification network is used in the project.
By training this convolutional neural network, the network will have the energy to extract the features of the game screen. With these, the algorithm can generate the next operation instructions.
As shown in the picture:
intelligent-uavpath-planning
Gitee：https://gitee.com/wwy2018/intelligent-uavpath-planning-simulation-system-S
intelligent-uavpath-planning - This project is an intelligent UAV path planning simulation system
This project is an intelligent UAV path planning simulation system, which has the advantages of fine operation control, strong platform integration, omnidirectional model establishment and application automation.
It is based on the fact that two countries, A and B, are launching a UAV war in Area C. The core function of the system is to plan the UAV route through the simulation platform and verify the output. The data can be imported into the real UAV, so that it can accurately arrive at any position on the battlefield according to the prescribed route, and support multi-person and multi-equipment formation joint operations.
As shown in the picture:
EssayKiller_V2
Github(4.9k)：https://github.com/EssayKillerBrain/EssayKiller_V2
EssayKiller_V2 - One person used three months to develop an artificial intelligence that can write essays
Generally speaking, the composition of the college entrance examination can achieve a satisfactory score through batch and model training. In fact, it is to train students as a writing machine, input a large number of sample essays for students, provide writing modes, and then study for a long time. Wait, isn't this the same as the training process for artificial intelligence?
So, if an artificial intelligence system is given the same data, can it be taught to write essays?
The answer is, of course. The UP master of station B - Turing's cat, adhering to the concept of action is worse than action, one person used three months, 500 hours, 10,000 lines of code, 200000000 pieces of data, 1700000000 parameters, and developed an artificial intelligence that can write essays - Essay Killer.
As shown in the picture:
style2paints
Github：https://github.com/lllyasviel/style2paints
style2paints - No need to install any environment, do any configuration, download and double-click, you can color the line drawing.
This AI project is the gospel of the second dimension. Although it is driven by AI, this project does not require you to install any environment or do any configuration. Download directly, and then double-click to color a line drawing.
As shown in the picture:
