CNN Receptive Field | Deep Learning Animated

Поделиться
HTML-код
  • Опубликовано: 24 ноя 2024
  • НаукаНаука

Комментарии • 31

  • @looooool3145
    @looooool3145 5 месяцев назад +16

    this video is a hidden gem! may you get blessed by the algorithm soon!

  • @TBBots
    @TBBots 5 месяцев назад +5

    Amazing, nailed the 3Blue1Brown style!

  • @macegamma431
    @macegamma431 2 месяца назад +1

    Incredible explanations!!! Wow the way you’ve condensed the information and then packaged it with the lovely animations to send it straight to intuition space :)

  • @academicendeavour3655
    @academicendeavour3655 2 месяца назад

    This was a great delivery of the subject , thank you!

  • @jakubner8824
    @jakubner8824 3 месяца назад

    I'm amazed by the quality of this tutorial!!!

  • @donk9613
    @donk9613 4 месяца назад

    This is the most underrated channel i found

  • @sharjeel_mazhar
    @sharjeel_mazhar 3 месяца назад

    A video of all of the top CNN based CV models would be great, like Alex net etc

  • @DavidW.-is3wb
    @DavidW.-is3wb 5 месяцев назад +1

    Underrated treasure! Guess math formulas on the cover really scare people away?

    • @Deepia-ls2fo
      @Deepia-ls2fo  5 месяцев назад

      Thank you ! I'll try not to include them in the next thumbnail aha

  • @yongwookim1
    @yongwookim1 3 месяца назад

    Excited to see whats coming next😮😮😮

  • @Bwaaz
    @Bwaaz 3 месяца назад

    amazing quality thanks !

  • @swagatbaruah522
    @swagatbaruah522 Месяц назад

    Thanks man.
    But I'm want to point something out. I might me wrong, but:
    After passing the image through the 3 layers, each layer reduces the size of the image (width, height) of the feature maps, which are not shown here. In the video, you didn't show the feature maps getting smaller.

    • @Deepia-ls2fo
      @Deepia-ls2fo  Месяц назад

      @@swagatbaruah522 Hi ! The layers do reduce the size of the features maps if you don't add padding. Here you can assume that everything is zero-padded to preserve width and height. :)

  • @ocamlmail
    @ocamlmail 2 месяца назад

    Thank you, very clear!

  • @CharlottePecolo
    @CharlottePecolo 5 месяцев назад

    Very nice video !

  • @anapauladesousa5701
    @anapauladesousa5701 2 месяца назад

    Just amazing! 🥰

  • @0810signor
    @0810signor 4 месяца назад

    What if you use dilated? How will the effective receptive field change?

    • @Deepia-ls2fo
      @Deepia-ls2fo  4 месяца назад

      @@0810signor things get a little weird with dilated convolution ! The receptive field increases faster but is more sparse

  • @jakubner8824
    @jakubner8824 3 месяца назад

    I try to reproduce visualisation of effective receptive fields 8:15 I am using TF and Torch, but don't know how to manually back propagate the output with one white pixel. Would You like to share the code or give me a hint?

    • @Deepia-ls2fo
      @Deepia-ls2fo  3 месяца назад +1

      Hi thank you for the kind words !
      Unfortunately I'm working on the next video and need a bit of time to clean up the code before publishing it on github. However, I created a small gist for you with the notebook I made to create these specific visualisations, enjoy. :)
      gist.github.com/Tmodrzyk/be07c3f91fbabf3aebb3e6ff0d3971ac

    • @jakubner8824
      @jakubner8824 3 месяца назад +1

      @@Deepia-ls2fo OMG I love You so much and keep my fingers crossed for your channel!!!

    • @voncolborn9437
      @voncolborn9437 3 месяца назад

      I'm anxiously waiting for your next videos. Would you mind sharing a little as to how you actually generate the graphics for your videos? How long does it take you to produce one of the 10-15 minute masterpieces?

    • @Deepia-ls2fo
      @Deepia-ls2fo  3 месяца назад

      @@voncolborn9437 Hi thanks for the kind words, don't be anxious though aha !
      I mainly use Manim, a little of Blender for the 3D animations and then video editing with Da Vinci Resolve.
      It takes around 60 hours to produce a 15min video. Hope this helps !