Author Topic: Questions about CNNs  (Read 1547 times)

infurl

  • *
  • Autobot
  • ******
  • Posts: 221
  • Humans will disappoint you.
    • Home Page
Re: Questions about CNNs
« Reply #30 on: December 20, 2016, 09:17:31 PM »
@korrelan say yes, it's easier

LOCKSUIT

  • *********
  • Terminator
  • Posts: 880
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: Questions about CNNs
« Reply #31 on: December 20, 2016, 09:21:58 PM »
I've went and studied them many times and just yesterday a 3 page all about them padding relu pooling dropout stride filter amount and size and layer amount.

His post sounding nothing like them. He kept talking about anything but my question.

korrelan

  • *
  • Replicant
  • ********
  • Posts: 527
  • Look into my eyes! WOAH!
    • Main YouTube
Re: Questions about CNNs
« Reply #32 on: December 20, 2016, 09:27:17 PM »
I tried... yes lock...

@ Infurl  :)
It thunk... therefore it is!

LOCKSUIT

  • *********
  • Terminator
  • Posts: 880
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: Questions about CNNs
« Reply #33 on: December 20, 2016, 09:31:09 PM »
The only part that finally but breifly says the answer is:

Quote
The CNN never has to store whole images; using just eight feature maps (more is better) you can describe thousands of objects. You only need to store the feature map numbers and their relative positions to each other.

So I'm guessing 7M mems & only 1,000 filters OK.

But does 1,000 filters allow any SINGLE *1 of the 7M images to be selected? Or 1,000 filters means only 1,000 possible selections?

infurl

  • *
  • Autobot
  • ******
  • Posts: 221
  • Humans will disappoint you.
    • Home Page
Re: Questions about CNNs
« Reply #34 on: December 20, 2016, 10:06:29 PM »
@locksuit

start writing the software and you'll soon figure it out

parameterise everything so you can try different values to find the ones that work

use Python, it has training wheels

and FFS stop annoying the grown ups

LOCKSUIT

  • *********
  • Terminator
  • Posts: 880
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: Questions about CNNs
« Reply #35 on: December 20, 2016, 11:04:31 PM »
I cannot write what I don't understand.

kei10

  • It's a honor to meet everyone!
  • *
  • Starship Trooper
  • *******
  • Posts: 441
  • Just kidding.
Re: Questions about CNNs
« Reply #36 on: December 21, 2016, 01:38:02 AM »
First, start writing basic neural network before working your way towards harder ones like CNN, and then you will get everything even easier rather than asking questions again and again and still doesn't understand.

What do they say? First time for everything, and baby steps.



Expected reaction from Locksuit: "Hell no, I am the most intelligent being ever, oho, oho, oho, oho, oho." ... Then he proceeds to ask more questions.

Here, in case you're too lazy to get python, let me put it here so it will be a life saver for you.

https://www.python.org/
Greetings, signature.

LOCKSUIT

  • *********
  • Terminator
  • Posts: 880
  • First it wiggles, then it is rewarded.
    • Enter Lair
Re: Questions about CNNs
« Reply #37 on: December 21, 2016, 02:08:25 AM »
Quote
I cannot write what I don't understand.

I meant CNNs.

keghn

  • *
  • Starship Trooper
  • *******
  • Posts: 304
Re: Questions about CNNs
« Reply #38 on: December 21, 2016, 05:37:12 PM »
 Well Cnn is a big detector neural network. But what it really is a tiny nn doing all the work. Instead of having a big input
layer that take in a image of 1600 x 900. A cnn uses a tiny NN of 10 x 10 that scans the whole picture. The little
networks travel over the image it look for lines, corner lines, and other lines. When it over one it activates
and sends it forward to a empty image that record the Little scanning NN output.
 Once done, then a new little NN scans the information collected in the second level. The little scanning NN look for combination of line together that make up a finger or an
eye. These detection are past to a new empty image, the third layer. That capture the output of The tiny scanning NN.

On the third layer. The third level scanning nn on level look for combination of detection found on the second level. like faces, arms,
and chest and stomachs. Next layer is people, cars building and so on. Next lay is mountains, moon, and forest. Next
layer is galaxies, nebulas, and constellations.

 

Welcome

Please login or register.



Login with username, password and session length

Users Online

25 Guests, 2 Users
Users active in past 15 minutes:
LOCKSUIT, squarebear
[Trusty Member]
[Terminator]

Most Online Today: 47. Most Online Ever: 208 (August 27, 2008, 08:24:30 AM)

Articles