hi... thanks for the vdo, bt can I use it for a set of 200 images or more? and one more thing, when both original image and super resolution images are similar ,what is the need of super resolution?
Yes, absolutely. You can use it for a set of 200 or more given thats your testing set. For training, we need a much larger set and we are using pre-trained weights for the sake of avoiding high gpu consumption that training typically takes. The SRCNN model is a basic model with just 3 neural network layers. In the world of image super resolution, there are way more deep neural networks with a huge number of layer, way more powerful models that result in astonishingly sharp changes on low resolution images. This model is usually used to learn to fundamentals and understand how image super resolution works. The intention is not to use it to actually perform super resolution on real images captured by our cameras with huge number of pixels. But for less number of pixels, on standard test set, the model performs well even on visual levels. Even, google phones, use image super resolution as post processing to produce picture qualities which are impossible to capture without dslr camera like huge lenses. But the models they use surpasses this basic model that we are studying as a part of research by leagues. However, like any technology, the fundamentals stays the same and without them you cannot comprehend the versions built on top of it. Thus, from study point of view, research point of view, this is an important model, but of little use in actual image super-resolution tasks on pictures captured by cameras. I hope this generates clarity over your concerns, thanks!
Hi @Lachu-hi6gg, Sure, here is what I am aware of: EDSR, WDSR, and SRGAN: ◦ The GitHub repository provides implementations of EDSR, WDSR, and SRGAN for single image super-resolution. You can load pre-trained weights and apply super resolution to images 2. ◦ Check out the code and examples in the repository here. github.com/krasserm/super-resolution
@@My-bself I see, that should not be happening. It does download for me, can you try this link in that case, drive.google.com/drive/folders/1e5u38BG6lBjUm24uRb4WS3lmms0r6UtM?usp=share_link
Most welcome! About creating your own h5 file, I am afraid I could not do it either. I observed that google colab is not enough to train SRCNN model and my computer has neither the amount of GPU that training requires. I am afraid I cannot help you here. Thanks.
Hi @saurabhsingh5948, I see that you are looking to creating your own h5 file rather than using one. I am afraid but I could not do it either. I observed that google colab is not enough to train SRCNN model and my computer has neither the amount of GPU that training requires. Thank you.
Hi @ramakrishnamiryala, The link does not open a webpage but instead downloads the zip directly through your browser. Can you please check if your browser if blocking the download or try a different browser. Thanks.