CNN(Convolutional Neural Network) Visualization

preview_player
Показать описание
I had the wonderful opportunity to showcase my work at Design Korea 2024 under the name 'Neural Network'.

Previously, I exhibited the same project at CVPR art gallery, but this time, it has been further upgraded.

I implemented everything including tensor operations, activation functions, tensor visualization, and so on using Processing.

The entire process and all the code are available on my blog and github as open source!
A heartfelt thank you to everyone who supported me and visited the exhibition!

Рекомендации по теме
Комментарии
Автор

Didn’t know what the hell it was, but watched the whole thing because it was satisfying

mindhooked
Автор

My brain when the waiter asks me how many scoops of ice cream I want

sockatume
Автор

for anyone wondering, the ai doesnt actually take that long to recognize the number, its just giving a visualization of the neural network and each layer of the NN
edit: didn't expect this to blow up, my explanation is not very clear, I'd recommend searching up a 3b1b video on how neural networks work if you're interested

gtALIEN
Автор

작품설명

<인공 신경망>은 인공지능의 작동 원리를 예술적 표현을 동원하여 구현한 인터랙티브 미디어 작품이다. 이 작품은 복잡한 AI의 계산 과정을 빛과 소리로 바꿔 그 과정을 관객이 직관적으로 이해할 수 있도록 고안되었다. 데이터가 신경망을 통해 전달, 학습되는 과정을 시각적, 청각적으로 표현하여 관객이 AI가 실시간으로 정보를 처리하고 반응하는 모습을 직접 체험할 수 있다. AI가 만들어내는 이미지와 영상이 점점 더 많아지는 가운데 <인공 신경망>은 AI의 기본 원리가 사칙연산 같은 간단한 계산에서 시작된다는 점을 알기 쉽게 보여준다. 관객이 AI의 추론 과정에서 실제로 어떤 일련의 과정이 벌어지는지를 확인할 수 있게 돕는 이 작품은 동시에 AI가 우리의 일상에서 어떤 역할을 하고 또, 앞으로 어떻게 더 진보할지에 관한 궁금증을 자극한다.

심심-yb
Автор

This is really cool, the video is a representation of how the computer is sorting through the data (pixels turned on or off 1, 0), the code itself isn’t terribly complex either. The real challenge is getting it accurate more often than not

abouthobos
Автор

I like how in the middle it turns it to a generic 3 as an internal understanding of the 3ness of the image.

gljames
Автор

This is what my brain does whenever my wife asks “what did I tell you to do this morning?!”

Sevenseas
Автор

I think, we already understand our mind fully. we wish for more depth, more space, we have reached a box outside of thought and now walk in circles having forgot

milesbismarck
Автор

A geat representation of the basic CNN Functions
1. Convolution(with filters/kernels)
2. Pooling
3. Flattening
4. Fully connected layer
There is 2 convolution and 2 pooling layers I think followed by flattening and prediction. (Knowledge from studying for the CNN Exam 2 days ago)
PS: Pls feel free to correct me if I am wrong as I'm amatuer student as well.

beckhamroshan
Автор

Man, that's pretty dope illustration of convolutional neural network, common deep learning algorithm. In this case for alphanumeric character recognition. Today's job for AI or ML engineers

bawonos
Автор

Guys your code is very cool, I have been doing research on deep learning vision for many years but your animation effect still shocked me

格瓦拉窃-sh
Автор

This is best explained CNN (convolution neural network)
This is use for image recognition in deep learning 😮❤❤

Muhabullah-yb
Автор

The thing is, even you draw a random stuff, it will still predict a number for you.

xqz
Автор

They nailed the naming when they used "convolutional" jesus f I'll never get that time back

charleslogan
Автор

-Какая это цифра? 😄
-Эмм.. (Завис на 3 минуты) три?!! 🧐

МаксимЭлектрик-ры
Автор

An explainer of what it’s doing would be cool for those of us without doctorates lol

ActualLiteralKyle
Автор

Вы тоже зашли в комментарии узнать, что там черт возьми, вообще происходило и какая была конечная цель этой длительной анимации?

LethalMelody
Автор

its fine we give input. we got it all ready

JarppaGuru
Автор

"AI will revolutionize the tech sector"
Meanwhile, the AI - "After 4 years of research, having exhausted my funding twice, and at the cost of countless lives, I believe the number you showed me was, in all likelihood, a three."

swguygardner
Автор

As i understand it, watching it create the conditions for such a number to arise, and referencing it's dataset to identify it, is elegant and beautiful

DanVogt
visit shbcf.ru