SRGAN Implementation on Custom dataset | Super Resolution GAN

[@HariharanS-bu5jz]

Thanks for this video, for me this project is a nostalgia of my college days....

[@CodeWithAarohi]

Glad you enjoyed it!

[@kanugondarohithkumar300]

Brother can you share ur contact details. I am working on this project only. So i have doubts

[@noorahmadharal]

It was very helpful video.... Thank You, ma'am

[@CodeWithAarohi]

Most welcome 😊

[@kvaibhav23]

this is the video which I was looking for since so long from you.

[@CodeWithAarohi]

And finally I made it 😊

[@81_saibalbera68]

Thank you for this informative video. But can also show a video where you can show how to deploy the model . And also how did you made the size of images all same ?

[@CodeWithAarohi]

Noted!

[@XavierCliment]

Thanks for this wonderful tutorial, is it possible to continue training with models that you have previously trained? Continue training from a model, thank you very much

[@CodeWithAarohi]

Yes, absolutely

[@XavierCliment]

@@CodeWithAarohi great! What parameters I have to put? thanks

[@Dinesh-wc8jw]

mam,i trained the model upto 8000 pretrained epochs succesfully,after starting fine tune epochs it throw's an value error. ValueError: Using a target size (torch.Size([16, 1])) that is different to the input size (torch.Size([8, 1])) is deprecated. Please ensure they have the same size. could you please help me mam....

[@mumbaikachallenge5526]

This is great explanation❤️

[@CodeWithAarohi]

Glad it was helpful!

[@minthway2736]

Please ! Can you do ESRGAN training on a custom dataset? I tried training ESRGAN on a custom dataset, but this did not work, and I was looking for related videos on KZbin, but this did not work. Could you please help me? Thank you a lot!!

[@kanikabisht6331]

Thank you very much mam for such a wonderful video..

[@CodeWithAarohi]

Most welcome

[@Shashank_Shahi1989]

Your video's are awesome. I don't know why you have so much low views. If it's possible from your side can you explain Transformers in detail and how to Implement PaLM in reality to create chat bot like ChatGPT ? Any guide or books or step by step guide ? Can you explain Reinforcement Learning From Human Feedback ((which ChatGPT use internally )) ? Thank You

[@abubakarsaleem5167]

Maam how we resume the training, if my training got stopped

[@usmanzafar6817]

hello, your videos are very helpful. kindly share some example/video on super resolution of medical images (MRI, CT Scan etc). thanks

[@CodeWithAarohi]

Will try

[@MohamedElsayed-bm5qw]

First of all, thank you very much for such an amazing video. I want to upscale 64*64 images to 128*128, what is the required number of epochs and how much time each epoch might take?

[@CodeWithAarohi]

The number of epochs required for training a SRGAN can vary depending on several factors but to get a rough estimate, you can start with a small number of epochs and monitor the performance of your model on both the training and validation sets. If the performance is improving and not yet plateauing, you can continue training. If you notice that the performance on the validation set is not improving, or if there are signs of overfitting, you may want to stop training.

[@MohamedElsayed-bm5qw]

I run the training command on a 250 images dataset for almost 12 hours now but there is no sign of training and no output (only * sign beside the cell) is it okay? I also want to change the number of epochs and upscaling factor and trace the model performance can you tell me how to do that? @@CodeWithAarohi

[@KhushbooKhushboo-n1l]

hello mam, can we do like we have 25 images of same target but from different angle. i want one high resolution image from this. is it possible

[@SaimKhalid-wx2zv]

Thank You so much, always learning a lot from you

[@CodeWithAarohi]

Welcome 🤗

[@kilykk]

It is possible to use this code with grayscale images?

[@harharmahadev8624]

Thanks u so much ma'am, I was waiting for this 🙏

[@CodeWithAarohi]

Welcome 🤗

[@palakjain521]

hey, can we use this code for mGAN Super resolutio too?

[@hansmueller7349]

Very good video! I found a small bug. If you want to change the scaling, you have to adjust the variable scale in the file srgan_model in the class Generator. I've got one more question. The script always keeps the form. Can you change the script so that shape changes are also learned? e.g.: i take pictures of old car models and pictures of the new car models that go with them. Then the AI can take any car and use it to calculate a newer model

[@pavinderyadav4654]

Hello, ma'am! It's an extremely informative video. Can you recommend some super resolution models that can be used in real-time applications?

[@CodeWithAarohi]

ESRGAN (Enhanced Super-Resolution Generative Adversarial Network), SAN (Second-order Attention Network), RCAN (Residual Channel Attention Networks)

[@pavinderyadav4654]

@@CodeWithAarohi Thanks for your reply Ma'am. Can you make a tutorial on how to use any of these technologies, to work with any object detection algorithm like yolov8, to enhance the accuracy of the detection.

[@mumbaikachallenge5526]

Greatest! Amazing

[@CodeWithAarohi]

Thanks a lot!

[@cemrealdogan7163]

Hi Aarohi, is it possible to modify your py script for making SRGAN network RS-ESRGAN (Remote Sensing Enhanced Super Resolution GAN)? I heard RS-ESRGAN is more effective for satellite images.

[@CodeWithAarohi]

Yes, it is possible to modify the py script for making the SRGAN network into an RS-ESRGAN network. RS-ESRGAN is an enhanced version of ESRGAN, which is a type of GAN specifically designed for super-resolution of images. The main difference between SRGAN and ESRGAN is that ESRGAN uses a more advanced generator architecture, which includes residual-in-residual dense blocks (RRDBs) and a feature fusion module (FFM). RS-ESRGAN, on the other hand, is an extension of ESRGAN that incorporates remote sensing-specific modifications to the network architecture and training process.

[@CodeWithAarohi]

To modify the SRGAN py script for RS-ESRGAN, you would need to implement the following changes: Replace the generator architecture: The generator architecture used in SRGAN is based on a deep residual network (ResNet) with skip connections. To implement RS-ESRGAN, you would need to replace this architecture with the one used in ESRGAN, which includes RRDBs and an FFM. Add remote sensing-specific modifications: RS-ESRGAN includes several modifications to the network architecture and training process that are specific to remote sensing applications. These modifications include using a loss function that takes into account the spectral and spatial characteristics of the input and output images, as well as a multi-scale training strategy that trains the network at different resolutions. Train the network: Once you have modified the network architecture and training process, you will need to train the network on a dataset of remote sensing images. This may require collecting or acquiring a suitable dataset of remote sensing images, as well as fine-tuning the network's hyperparameters and training settings to optimize performance on this type of data.

[@cemrealdogan7163]

@@CodeWithAarohi Thank you. It is much appreciated. 🙏🙏🙏

[@RADEONR]

So how to find a (software/platform) that can generate video into HR-LR frames or particle blocks of each frame/image as png /jpg

[@RADEONR]

Inshort how u create your dataset?? and i can do it with video by exctracting it frames?!

[@CodeWithAarohi]

I collected the high resolution images and then converted them into low resolution images. Yes, you can extract frames

[@cemrealdogan7163]

Hi Aarohi, can I use images at .tiff format instead of png for inputting dataset folder? Thanks.

[@CodeWithAarohi]

I never tried it. But I think with few modifications we can do that.

[@taneyasoni]

where to get this cars dataset from

[@CodeWithAarohi]

You have to prepare it yourself.

[@taneyasoni]

@@CodeWithAarohi So that means I have to collect HD photos and make low resolution from them?

[@alikassi-vc8xt]

Hello, where can I retrieve the image files for training and thank you

[@CodeWithAarohi]

You have to prepare your dataset yourself. Just collect some images from internet and then make them low resolution

[@spaceghost-fj8co]

is it possible to use SRGAN model on video dataset, if it is possible please provide some insight about it

[@CodeWithAarohi]

Yes, it is possible to use the SRGAN model on a video dataset to enhance the resolution of individual frames within the video. However, there are some additional considerations to keep in mind when using SRGAN on video data. One approach is to extract individual frames from the video and apply the SRGAN model to each frame independently. This can be done using a video processing library like OpenCV, which provides functions for reading video files and extracting frames. Once you have extracted the frames, you can pass each one through the SRGAN model to generate a high-resolution version of the frame.

[@263boi]

Hi i cannot seem to find the solution.pynb file

[@nitinujgare]

Me too...can anyone share that file if removed from github

[@bhanuprakashreddy]

Thank you

[@CodeWithAarohi]

Welcome!

[@zakwanibrahim7283]

you drag too much only last 5 min are valuable ... thanks

[@CodeWithAarohi]

Thank you for the feedback

[@taneyasoni]

no you do not. I really needed those explanations. thank you Aarohi@@CodeWithAarohi

[@KanugondaRohithKumar-h6r]

Ma'am I have doubts in environmental setup . could u please share ur gmail?

[@CodeWithAarohi]

[e-mail]