Inception V3 Practical Implementation | InceptionV3

[@ashv7055]

At 20:20 - In flow chart you have mentioned 3x3 layers but in code you are writing 5x5. Any reason for that?

[@CodeWithAarohi]

Join My Channel for Additional Benefits : Click on JOIN Button and choose the Membership Plan on the basis of benefits or Perks which I am offering to Members.

[@mahalerahulm]

Thank you Aarohi Ma'am. you are making it understand such complex architectures easily, and also showing it practically it is really very great job , hats off !!

[@CodeWithAarohi]

Glad you think so! Keep Watching 🙂

[@andreramos3290]

good job. Very nice. continue with this teaching work.

[@CodeWithAarohi]

Thank you, I will

[@dineshvennapoosa4072]

Thank you ma'am for this, it really helped me a lot in understanding the CNN architectures.

[@CodeWithAarohi]

Glad my videos are helping you

[@josephorimolade5626]

An excellent and brilliant presentation. Your videos have given me a better understanding of the CNN architecture. However, I would like to request that you make a video to complete the implementation of the Auxiliary part as well as InceptionV4. Secondly, I also look forward for your video on ResUNet. Thank you. Joseph

[@CodeWithAarohi]

Glad my video is helpful and surely I will make video on requested topics

[@newsanalysis8789]

maam isme model summary ke baad epochs nhi chalenge kya? epochs chalane ke liye kya likhna hoga

[@vi_shw]

I just found a perfect code implementation of Inception V3.

[@CodeWithAarohi]

Glad video is helpful 😊

[@riyaroyr.r6964]

Thank you madam for sharing this, great explanation

[@CodeWithAarohi]

My pleasure 😊

[@QubitBrain]

Why block A in diagram have convulation layer of filter size 3x3, when in code you have written filter size of 5x5?

[@sihamberkat8347]

how can i implement inception v3 using digital data (not images) for prediction ?

[@angelinvideos]

I just found a perfect code implementation of Inception V3. You've nailed it. I have a question in the code at 32:58. I do not see the dense layer with 2048 neurons in the code just before the final dense layer with 1000 output. Could you please clarify?

[@manishsingh-vk8if]

She has forgot to add it. Also no implementation of auxiliary layers.

[@osamanusrat2404]

thanks for very good explanation can u please elaborate how did you come to know the no of filters we are using in inception block and reduction block. Rest explanation is perfect. Thanks alot

[@VikasKumar-ci6ps]

You are my favourite teacher

[@CodeWithAarohi]

Vikas Raj thankyou vikas

[@atenaseifi8099]

Thank you very much for the Inception tutorial series. I still don't know how can I train my own data with these layers? I have already trained the model with the Cifar10 dataset but I couldn't find an explanation about adding custom data to Inception, like what you said in the EfficientNet video. I will appreciate it if you can help.

[@sitarammeena2038]

thank you mam, its great effort by you . very clear explaination .

[@CodeWithAarohi]

It's my pleasure

[@hamzabiyuzan6969]

hey first of all thanks a lot for your videos they are all amazing, just a question , in the function inc_block_a() for the branch5x5 the size of the second layer is 3x3 not 5x5 because the layer with the size 5x5 got replaced with 2x 3x3 size layers or I just didn't well understand ?

[@CodeWithAarohi]

Glad my video helped you and yes you are correct with with your understanding.

[@ashv7055]

Everytime we are using different filters like 64, 96, 384 etc. Do we need to remember all these filters?

[@CodeWithAarohi]

No need to remember.

[@ijkim19]

So what would you do if you didn't have the weights file to import?

[@CodeWithAarohi]

You can work without weights also

[@rajamurishiva2084]

why did u call inc_block_b 4 times and inc_block_a 3 times ...? calling them once is enough to get inception block

[@VikasKumar-ci6ps]

Very knowledgefull video

[@CodeWithAarohi]

Thanks

[@pavithramuthukumar3802]

Generally the medical images are grayscale images.so,how can we train a graysacle image to a inception v3 model.Kindly give your suggestions regrading this.Thanks for your contribution....

[@CodeWithAarohi]

Suppose the image is grayscale then Image size would be suppose 300*300*1 here 1 is a number of channels for greyscale Image. So we just changed the number of channels from 3(RGB) to 1(greyscale). Rest of the code would be same.

[@pavithramuthukumar3802]

@@CodeWithAarohi thankyou so much will try the same

[@CodeWithAarohi]

Welcome 😊

[@RenukaAIChronicles]

Hello madam,How to convert a 32*32 image to 299*299 , as it is required for CIFAR 10 dataset , since the inception takes only 299*299.

[@CodeWithAarohi]

You can use opencv2 for resizing but you will get blurry looking image.

[@agnishpaul8683]

Mam where can I find the code?

[@CodeWithAarohi]

github.com/AarohiSingla/InceptionV3

[@agnishpaul8683]

Thanks mam@@CodeWithAarohi

[@NIKHATPARVEEN-ud2tj]

How the no. filters are been generated like 32, 64, 80

[@CodeWithAarohi]

The number of filters in each layer is a hyperparameter that is chosen by the network designer based on factors such as the complexity of the task, the size of the input data, and the available computational resources. Generally, the number of filters increases as we move deeper into the network, as this allows the network to capture more complex features at higher levels of abstraction. In the case of InceptionV3, the number of filters in the first convolutional layer is 32, while the number of filters in subsequent layers varies between 64 and 320. These values were chosen based on experiments conducted by the authors of the InceptionV3 paper, which showed that these values provided a good trade-off between model complexity and performance on image classification tasks.

[@hectorramos5489]

Do you have the notebook? Thank you

[@CodeWithAarohi]

Yes i have. Will share github link tomorrow

[@CodeWithAarohi]

github.com/AarohiSingla/InceptionV3/blob/main/inceptionV3_implementation.ipynb

[@subashchandrapakhrin3537]

Good Video

[@CodeWithAarohi]

Thanks

[@KOTESWARARAOMAKKENAPHD]

can you please provide code link

[@CodeWithAarohi]

github.com/AarohiSingla/InceptionV3

[@RenukaAIChronicles]

Hello madam, is it possible to share the codefile, it will be great.

[@CodeWithAarohi]

github.com/AarohiSingla/InceptionV3

[@shilpashreehs4868]

Thanks a lot

[@CodeWithAarohi]

Most welcome

[@anjithaanju2584]

Can you make a video about Inception v4?

[@CodeWithAarohi]

will do that soon

[@anjithaanju2584]

@@CodeWithAarohi thankyou mam

[@dashbiplob5351]

mam! where is code?

[@CodeWithAarohi]

github.com/AarohiSingla/InceptionV3

[@puspabenmodi4638]

Can you send this code file

[@CodeWithAarohi]

github.com/AarohiSingla/InceptionV3/blob/main/inceptionV3_implementation.ipynb

[@MrRabeaahmed]

why did you still write in 'branch5x5 ' 5X5 instead of 3x3? for my understanding from your previous video block A change 5X5 convolution to 3x3 convolution 2 times I think that there is a mistake in code. my correct code: def Inception_Block_A(x): Conv1x1 = Conv2d_B_R(x, 64, 1, 1) #64 filters of size 1X1 Conv3x3 = Conv2d_B_R(x, 48, 1, 1) #48 filters of 1x1 Conv3x3 = Conv2d_B_R(Conv3x3, 64, 3, 3) Conv5x5 = Conv2d_B_R(x, 64, 1, 1) Conv5x5 = Conv2d_B_R(Conv5x5, 96, 3, 3) Conv5x5 = Conv2d_B_R(Conv5x5, 96, 3, 3) Conv_pool = AveragePooling2D((3,3), strides=(1,1), padding='SAME')(x) Conv_pool = Conv2d_B_R(conv_pool, 32, 1, 1,) x = layers.concatenate(Conv1x1, Conv3x3, Conv5x5, Conv_pool, axis=channels_axis) return x PS: Please correct me . with my respect for you

[@yogov3012]

Did you changed the code by your own? Did you referred any other articles for InceptionV3?

[@Tanmay_Bhattacharya]

I guess MaxPool2D should have been used as presented in the architecture diagram. Rest of the part of your code is correct according to the diagram.

[@zehra2334]

Your videos and explanations are nice but your microphone is very bad :(

[@CodeWithAarohi]

Sorry for inconvenience. I have changed the microphone in my recent videos :)

[@israanasir4022]

please arabic language

[@alis5893]

I think at about minutes 20-23 you keep talking about a 5x5 convolution, but the image you show for the inception net is only showing 3x3 . Perhaps you made a graphical error. However your videos are amazing. Thank you

[@CodeWithAarohi]

My bad