Hello Sreeni I am working on 900x1200 dehaze multiview images uisng pix2pix GAN ,However the model needs input size of the images to be 256x256 . So how to handle non squared images divided into 256x256 patches
I am getting good results while testing from one of the trained samples. Although when I test the model with an entirely new image (one that is not trained as a part of the model), the results are not so convincing. A similar looking image is generated irrespective of the input image. Does anyone know how to solve this?
I have a Corona image, only in PAN band, and I would like it to be colorized which is now big issue for me, because there is no colored image to use asa training image. So can I use any PAN image or multiband image from recent Landsats? Thank you
Can we use GAN on sound with respective to nyquist condition, so neural network will create new sound using our sound dataset. there's a special algorithms for audio but i wanted to make it with image GAN in both we are dealing with number. Is it possible that GAN generated data to be reconstruct in expected output
Sir one question, we've downloaded the dataset as a zipped file, and on using that downloaded folder as the path in the code provided, we've been getting an error that says the system cannot find the specified path. How can we overcome this problem?
Thanks for the detailed explanation. Correct me if I am wrong, Your model already started producing very good results only after 10 epochs because you are testing on an input image which was part of training set. I wonder what happens if you use an input image which was not part of training set.
Great tutorials! I am just finishing up my undergrad in Geomatics Engineering. I was wondering if you could make some videos or point me in the direction of: -Adding external data such as radar readings to each pixel to make better use of all spatial data that is available to the user, not just pixels. Thanks tons!!
Thank you for this channel, it helps a lot! I have one question, if the training stops, how to load the saved model and restart training where it was let off?
Thank you for the fruitful video, but when i run the code on the last part I get this error ValueError: When there is only a single output, the `loss_weights` argument must be a Python float. Received instead: loss_weights=[0.5] of type , I tried with jupyter notebook and spyder, with both tiff and png images but i cant get to train , i get the same error. please assist.
i was having same errors, but then i removed loss weights from everywhere ( i'm running on kaggle), this issue got reolved, but now i am facing with some other issue.
Thank you for all your great videos ! I notice that you use Input from tf.keras.models while in previous video it was imported from tf.keras.layers. Is there a difference between these 2 imports ?
hello....your videos are amazing.... am very new to this Could you please let me know why this is happening...? cannot import name 'input' from 'keras.models how to do it Thank you
This video uses a custom data set, where we download locally and then load it. In my next video I will be using images I collected from a microscope. May be that is what you refer to as 'custom' dataset.
The code did not work with the keras == 2.10.0 version or higher. it worked with keras == 2.4.1 version. Thank you for the video. I am learnining a lot.
@joafin19 . I am experiencing problems running the code. Especially the training part there is confliction between weights=[0.5] and when I train it says it needs weights as a float. Is it the same problem you experienced?
I was waiting for this demonstration sir thanks a lot and keep supporting your virtual family 😊😊
Thank you so much 😀
Thank you very much for this amazing tutorial. This video has ironed out some of the things i was struggling with for px2pix + satellite imagery.
Glad it helped!
Hello Sreeni
I am working on 900x1200 dehaze multiview images uisng pix2pix GAN ,However the model needs input size of the images to be 256x256 . So how to handle non squared images divided into 256x256 patches
Thank you for sharing your treasure!!!
Why do you keep the BatchNormalization layers to trainable=True in the "define_gan" function?
Great work 👌👏
Thanks
Thank you very much. Please keep watching.
I am getting good results while testing from one of the trained samples. Although when I test the model with an entirely new image (one that is not trained as a part of the model), the results are not so convincing. A similar looking image is generated irrespective of the input image. Does anyone know how to solve this?
I have a Corona image, only in PAN band, and I would like it to be colorized which is now big issue for me, because there is no colored image to use asa training image. So can I use any PAN image or multiband image from recent Landsats? Thank you
Can the same model be used to colorize black and white images?
cannot import name 'img_to_array' from 'keras.preprocessing.image ..sir how to solve this error
Thanks sir but I have one problem related MAE value,where are we printing MAE value
Can we use GAN on sound with respective to nyquist condition, so neural network will create new sound using our sound dataset.
there's a special algorithms for audio but i wanted to make it with image GAN in both we are dealing with number. Is it possible that GAN generated data to be reconstruct in expected output
I am not much familiar with using GAN for audio inputs. Please check out WaveGAN, hopefully the right solution to your question.
Great tutorial sir! I've a doubt. Is this only the training phase? Where is the val folder used?
sir,can we use the same code for translating satellite images to optic images
thnks many sreeni
Sir one question, we've downloaded the dataset as a zipped file, and on using that downloaded folder as the path in the code provided, we've been getting an error that says the system cannot find the specified path. How can we overcome this problem?
I tried implementing this using UNET quite a while ago, wondering why using a UNET as generator never crossed my mind
One more request can you explain related to temporal consistency between multiview images with latest model ?
Amazing tutorial thank you sir keep up the good work
Glad you liked it
Thank you for your tutorial. Great.
You are welcome!
Thanks for the detailed explanation.
Correct me if I am wrong, Your model already started producing very good results only after 10 epochs because you are testing on an input image which was part of training set. I wonder what happens if you use an input image which was not part of training set.
Great tutorials! I am just finishing up my undergrad in Geomatics Engineering. I was wondering if you could make some videos or point me in the direction of:
-Adding external data such as radar readings to each pixel to make better use of all spatial data that is available to the user, not just pixels.
Thanks tons!!
Thank you for this channel, it helps a lot!
I have one question, if the training stops, how to load the saved model and restart training where it was let off?
Where can we use this concept? Please, Can anybody enlighten me?
Thank you for the fruitful video, but when i run the code on the last part I get this error ValueError: When there is only a single output, the `loss_weights` argument must be a Python float. Received instead: loss_weights=[0.5] of type , I tried with jupyter notebook and spyder, with both tiff and png images but i cant get to train , i get the same error. please assist.
i was having same errors, but then i removed loss weights from everywhere ( i'm running on kaggle), this issue got reolved, but now i am facing with some other issue.
Amazing tutorial, Thank You Sir
Sir could you also provide tutorials on Object Detection and Pose Detection using Keras.
Sir why you are focusing more on image related can we follow this series for learning ML????
Thank you for all your great videos !
I notice that you use Input from tf.keras.models while in previous video it was imported from tf.keras.layers.
Is there a difference between these 2 imports ?
They are all aliases, I should have stuck with one. I recommend using the ones from here: tf.keras.Input or tf.keras.layers.Input
hello....your videos are amazing....
am very new to this
Could you please let me know why this is happening...?
cannot import name 'input' from 'keras.models
how to do it
Thank you
Depends on your keras version Things can move so a quick Google search can help. ANyway, try: from tensorflow.keras.layers import Input
@@DigitalSreeni ... thank you so much....Sir
Am watching Your video from beginning....
So much information....
Thanks for the video, could you do an image to image translation with a custom dataset. Thanks
This video uses a custom data set, where we download locally and then load it. In my next video I will be using images I collected from a microscope. May be that is what you refer to as 'custom' dataset.
Hey! Your Videos are Amazing. Can you post some Videos On Diffusion.
Hello sir, Can this pix2pix model be used for multi-class image segmentation?
The code did not work with the keras == 2.10.0 version or higher. it worked with keras == 2.4.1 version. Thank you for the video. I am learnining a lot.
@joafin19 . I am experiencing problems running the code. Especially the training part there is confliction between weights=[0.5] and when I train it says it needs weights as a float. Is it the same problem you experienced?
how do i install this version of keras on kaggle, as it keeps showing dependency issue with tensorflow? what tensorflow did you use?