I understand that the image synthesis in Generative Adversarial Networks are capable of generating plausible images of some object. I have two doubts regarding this (I am new to this topic);
-Can this approach complete images that are incomplete/damaged?
-I understand that these images are just possibilities of what an object would look like, but what about the accuracy of the generated image, could an image be generated that is very close to the original complete image?
Error in image synthesis-Generative Adversarial Networks ?
Dear Ana Bucchi ,
Generative Adversarial Networks (GANs) have been widely used and it is expected to use for the clinical examination and image. The objective of the current study was to synthesize material decomposition images of bone-water (bone(water)) and fat-water (fat(water)) reconstructed from dual-energy computed tomography (DECT) using an equivalent kilovoltage-CT (kV-CT) image and a deep conditional GAN. The effective atomic number images were reconstructed using DECT.
https://www.sciencedirect.com/science/article/abs/pii/S001048252030442X
Regards,
Shafagat
https://www.jmlr.org/papers/volume23/21-0732/21-0732.pdf
Preprint Detecting GAN generated errors
There is an approach based on GANs for image inpainting, you can find it here: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8974211
GANs alone can only learn the implicit density function from the training dataset. One way to use GANs for image inpainting would be to train GANs with a complete dataset, then use the incomplete image known data as ground truth and run an optimization algorithm to find the closest generation to this target image. Hopefully, the generated image would present plausible content in the missing data region. Of course, this is just in theory, I think that the search would become prohibitively costly since the dimension of the latent space in GANs is typically 100 to 200.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training