Thak you for this explanation, I would like to learn more about the relationship between the gradient and the mask at the output, what I know, is that the gradient serves to draw points of mask at the output, then is there another relationship between the gradient and the mask?
Awesome video and very well explained. Maybe in the next videos, you should not switch between you and the architecture of the network, because that confuses me a little.
if you're referring to the number of channels that change with each level, they're associated with the number of filters you're using for the convolution operation - so each filter (from the convolution operation) will reduce the input size in the x/y dimensions (width and height -- given that it's an unpadded convolution) and will increase the number of channels by 1. for example if you look at the first block, your input image size is 572x572x1 (width x height x num_channels -- 1 for grayscale image, if you have RGB image, it's 3) -- with the first convolution operation (navy blue arrow to the right) a 3x3x64 convolution operation was applied -- meaning 64 different filters of 3x3 size were convolved with the original 572x572x1 image, resulting in an image with output size = [(572 + 2*0 - 3) / 1] + 1 = 570 for each dimension (unpadded convolution - and single stride - equation is (inputSize + 2*padding - filterSize / stride ) + 1 ) -- So you will get an output dimension of 570x570x64, which is what you see in the first convolution operation output in the diagram. As they mentioned, the number of channels are doubled with each level, meaning they increase the number of filters by a factor of 2 each time. You can check out Andrew Ng's video which explains this operation step by step (C4W1L08 Simple Convolutional Network Example on YT)
This video is very nice and well explained. very useful. will you please make a video in the topic W-Net: A Deep Model for Fully Unsupervised Image Segmentation
Thank you for the explanation. Helped a lot. Just a question if you could cover this topic in one of your videos- What is Anomaly detection?? Like for example detecting defects on surfaces... can u-net be useful there?
That is how you explain analytically, proper analytical approach, sad to see only 4.45K subscribers, dont worry i will share it
You are the best explicator I've seen so far. Thank you!
Hey… thanks for your kind words :)
Gruss aus Vancouver! Ausgezeichnet gemacht, klipp und klar, besten Dank!
Hey, danke dir ☺️
extremely dense and clear, thank you sir
Thank you very much :)
Thak you for this explanation, I would like to learn more about the relationship between the gradient and the mask at the output, what I know, is that the gradient serves to draw points of mask at the output, then is there another relationship between the gradient and the mask?
Greatly explained! Thank you. Can you make a video about RBMs and Deep Belief Networks
Nice and easy overview of Ronnebergers work
Thank you very much :)
You have a good explanation but your intro is tooooo long
Hey, thanks for your comment and feedback :)
Thank you Sir for the explanation.
too long intro. video starts at 1:50
Extremely Helpful thanks a lot
Hey man, thanks a lot :)
I got the basic intuition with this video. thank you.
Thank you for the explanation. Helped a lot.
Is it me or the volume of the video is lower than usual?
Awesome video and very well explained. Maybe in the next videos, you should not switch between you and the architecture of the network, because that confuses me a little.
Hey, thanks for your comment and your suggestion. I will try not to switch that often in the next videos :)
Hey may I ask what are these channels? I didn‘t understand that part.
hey man... could you provide me the time in the video of that part?
if you're referring to the number of channels that change with each level, they're associated with the number of filters you're using for the convolution operation - so each filter (from the convolution operation) will reduce the input size in the x/y dimensions (width and height -- given that it's an unpadded convolution) and will increase the number of channels by 1.
for example if you look at the first block, your input image size is 572x572x1 (width x height x num_channels -- 1 for grayscale image, if you have RGB image, it's 3) -- with the first convolution operation (navy blue arrow to the right) a 3x3x64 convolution operation was applied -- meaning 64 different filters of 3x3 size were convolved with the original 572x572x1 image, resulting in an image with output size = [(572 + 2*0 - 3) / 1] + 1 = 570 for each dimension (unpadded convolution - and single stride - equation is (inputSize + 2*padding - filterSize / stride ) + 1 ) -- So you will get an output dimension of 570x570x64, which is what you see in the first convolution operation output in the diagram. As they mentioned, the number of channels are doubled with each level, meaning they increase the number of filters by a factor of 2 each time.
You can check out Andrew Ng's video which explains this operation step by step (C4W1L08 Simple Convolutional Network Example on YT)
This video is very nice and well explained. very useful. will you please make a video in the topic W-Net: A Deep Model for Fully Unsupervised Image Segmentation
Thank you so much, I need next Example Project
Thank you for the explanation. Helped a lot.
Just a question if you could cover this topic in one of your videos- What is Anomaly detection?? Like for example detecting defects on surfaces... can u-net be useful there?
good information, thanks
Wow cool Video! Very well explained!
great explenation! thanks
You were so helpful🙌
Boah danke für dieses Video, rettest gerade meinen ass!!!!!
👍😂
thank you
please don't take too much time in intro
It is not clear whether this is a good idea to make a 0:54 second introduction I don’t think it is useful for you or anyone 😮😢
Fk this is a joke but it is an other introduction at 1:10 😮😢😮 without getting into it 😢😢😢
Then I got asked what I am waiting for 😂 1:33
I understood instantly why they called it a U net… if instantly implies that instant at the 3:00 mark when after waiting for 3 minutes I understood 😮
The music is bad
The idea of putting music is bad
I try to focus but it distracts me badly
Longest intro ever 🙄