TensorFlow 2.0 Tutorials for Beginners 7 - Use of Dropout and Batch Normalization in 2D CNN

Поділитися
Вставка
  • Опубліковано 24 гру 2024

КОМЕНТАРІ • 46

  • @WahranRai
    @WahranRai 3 роки тому +1

    38:18 You must REMOVE model = Sequential() from lines 5 and 9 `model already defined in 1 if you print your model you will check the architecture !

  • @KGPTalkie
    @KGPTalkie  5 років тому +1

    Hi, We are using 32 x 32 image size for faster training but you can use larger image size to bring the model accuracy past 80%. You can also try other tuning parameters to improve accuracy. Please comment below your accuracy.

    • @muhammadzubairbaloch3224
      @muhammadzubairbaloch3224 5 років тому +1

      Sir I think you are busy in some matter. But we are anxiously waiting for your videos

    • @KGPTalkie
      @KGPTalkie  5 років тому

      Thanks for encouragement. I am on leave. Will start LSTM next week.

  • @CarlosHernandezGarciadiego
    @CarlosHernandezGarciadiego 2 роки тому

    Once you trained the model at the end of the video, how do you select an image and a label and test the model with them to see if the model predicts well the animal?

  • @priyansurout19
    @priyansurout19 4 роки тому

    HI! I have question : Why we have to change this image to array form and not in minst fashion dataset. Help me out

  • @Justcaldo
    @Justcaldo 4 роки тому

    Hi KGP, I'm wondering why you use the test-set for the validation and not a true validation-set? Thanks! Great work

    • @KGPTalkie
      @KGPTalkie  4 роки тому

      Test set was unseen so it can be used

  • @makeoutloud
    @makeoutloud 5 років тому +2

    sir please make video on training on own dataset of like 4 or 5 different types and also show the prediction on live camera feed

    • @KGPTalkie
      @KGPTalkie  5 років тому +2

      Yeah sure. Please keep watching.

  • @inayatullah5576
    @inayatullah5576 5 років тому +4

    why are you initializing the object of sequential several times??

    • @KGPTalkie
      @KGPTalkie  5 років тому +4

      Opps. Thanks for pointing it out. That is a huge mistake. Sequential layer should be used only once before we start adding the layers. Thanks for suggestions.

    • @kevalrajpal4244
      @kevalrajpal4244 4 роки тому

      I was also wondering about the same issue!

    • @KGPTalkie
      @KGPTalkie  4 роки тому

      It was a mistake. I did copy and paste.

    • @kevalrajpal4244
      @kevalrajpal4244 4 роки тому

      ​@@KGPTalkie Sir I was following your series! Special thanks for the efforts, While following this video I was facing this error ValueError: Shape mismatch: The shape of labels (received (60,)) should equal the shape of logits except for the last dimension (received (20, 3)). Can you help me resolving this

    • @KGPTalkie
      @KGPTalkie  4 роки тому

      you need to reshape your data

  • @inayatullah5576
    @inayatullah5576 5 років тому +1

    if i use dataset of diabetic retinopathy and train it using VGG 16 model does it work??

    • @KGPTalkie
      @KGPTalkie  5 років тому +1

      Yes. It should work but before that please watch my videos on VGG16 network.

  • @RaushanKumar-qb3de
    @RaushanKumar-qb3de 3 роки тому +1

    Again Model.fit_generator is deprecated starting from tensrflow 2.1.0...☠️🙄 and i am using 2.6.0

  • @KGPTalkie
    @KGPTalkie  4 роки тому

    Thanks for watching. You can now Download Dog Cat Dataset by this link. github.com/laxmimerit/dog-cat-full-dataset

  • @m.revaldianggara7793
    @m.revaldianggara7793 4 роки тому

    hey do you help me? when ı use ImageDataGenerator(rescale=1./255) accuracy drops, normaly accuracy is very high when i'm using images range[0,255]. Why?

    • @KGPTalkie
      @KGPTalkie  4 роки тому

      Rescale is required only when pixel values are in between 0 to 255. Please check in your original data what is range of pixels.

  • @muhammadzubairbaloch3224
    @muhammadzubairbaloch3224 5 років тому +1

    please sir make video on LSTM, RNN etc

    • @KGPTalkie
      @KGPTalkie  5 років тому

      Yeah sure... From next week onwards, I will start with LSTM.

  • @prathameshjoshi1408
    @prathameshjoshi1408 4 роки тому +1

    I maybe late here but your github link is not working .. And since you have structured your data properly like train>dogs has dogs images and cats have cats images ! my dataset which is the same I downloaded from kaggle has all the images together so while image data gen It gives 37000+ images and you had 20000 images

    • @prathameshjoshi1408
      @prathameshjoshi1408 4 роки тому +1

      Btw Thanks u make awesome videos also u explain in a very great manner

    • @KGPTalkie
      @KGPTalkie  4 роки тому +1

      Thanks for watching. You can now Download Dog Cat Dataset by this link. github.com/laxmimerit/dog-cat-full-dataset

    • @prathameshjoshi1408
      @prathameshjoshi1408 4 роки тому +1

      @@KGPTalkie Thankyouu 💯..

  • @ihabshhadat134
    @ihabshhadat134 5 років тому

    good job but u have a single mistake in cell 23.
    you should remove lines 7 and 13 'model=Sequential()'
    I don't know how is it run with u

    • @KGPTalkie
      @KGPTalkie  5 років тому

      Thanks for watching. Yes that happened because I copied first three lines and forgot to delete first line during past. The github code is upto date. You can download and run it. It should work.

  • @jfabian
    @jfabian 5 років тому

    The variable "valid_data_dir" wasnt declared. Where should it point to?
    NameError Traceback (most recent call last)
    in ()
    ----> 1 validation_generator = datagen.flow_from_directory(directory=valid_data_dir,
    2 target_size = (32, 32),
    3 classes = ['dogs', 'cats'],
    4 class_mode = 'binary',
    5 batch_size = batch_size)
    NameError: name 'valid_data_dir' is not defined

    • @jfabian
      @jfabian 5 років тому +1

      I believe the correct variable is "test_data_dir"

    • @GilangD21
      @GilangD21 5 років тому

      @@jfabian yeah i think so too, but why in the video he can use valid_data_dir without problem? i didnt see any initialization with that name before

    • @jfabian
      @jfabian 5 років тому

      @@GilangD21 I'm sure he declared it before (not shown in the video)

    • @KGPTalkie
      @KGPTalkie  5 років тому

      Hi,
      It happened because jupyter notebook stores variables in cache and for some other project I had declared valid data dir and by mistake I used in this. You got it right, it should be test data dir

    • @jfabian
      @jfabian 5 років тому +1

      @@KGPTalkie good. Great work. Love your videos

  • @AdityaRaj-ix5rg
    @AdityaRaj-ix5rg 5 років тому

    valid_data_dir not declared??

    • @KGPTalkie
      @KGPTalkie  5 років тому

      Please change that to test_data_dir Thanks for watching.

    • @KGPTalkie
      @KGPTalkie  5 років тому

      The new corrected file is reuploaded. Please check and let me know if it is not working.

  • @kishanpandey4798
    @kishanpandey4798 5 років тому

    why is it 64, 128, 256 number of filters?

    • @KGPTalkie
      @KGPTalkie  5 років тому

      You can use any number of filters.

    • @kishanpandey4798
      @kishanpandey4798 5 років тому

      @@KGPTalkie How do we know which number to put?
      Can you prescribe a good book on Deep Learning? Thanks

    • @kishanpandey4798
      @kishanpandey4798 5 років тому

      @@KGPTalkie Please make a live project series from Web Data scrapping to Feature engineering- Model Selection-Evaluation-Deployment. It can help us get a good depth of how Actual Data Science and actual ML project works in an Industry and help us get industry insights

  • @RAZZKIRAN
    @RAZZKIRAN 3 роки тому

    very low voice,