[ad_1]
Nvidia revealed a new approach based on in-depth learning to eliminate noise, artifacts, text and similar undesirable aspects of photos. The Green Team researchers have collaborated with other researchers from Aalto University and MIT and will present this new grain removal tool at the International Conference on Plant Protection. machine learning in Stockholm, Sweden, later this week
. were trained by showing examples of noisy and clean image pairs. "Without being able to see what an image without sound looks like, this AI can remove artifacts, noise, grain, and automatically enhance your photos," the researchers claim. In addition, researchers say that this new approach requires less training time and can be faster in execution.
In the video above, you can see the AI denoising in action. This is worth looking at because artificial intelligence is being tested on various examples of photos and use cases. You will see the method of denoising based on deep learning removing noise of various types, from various inputs. The technique is not only good for old photos or those taken in low light, fast rendering, or MRI, it can also embellish corrupted images with text and shapes / blocks of colors superimposed.
GPU Nvidia Tesla P100 with the deep learning framework TensorFlow accelerated by cuDNN. 50,000 images from the ImageNet Validation Set were used in training.
I have been working in print for over ten years and the studios often provided very rough or suboptimal images. Maybe this Nvidia AI will be useful for this industry, where you often have to make the most of what you have received from a customer, but they expect that beautiful, accurate and accurate images leave the press. It certainly seems more useful than the standard denoising tools and related Adobe Photoshop tools I'm used to.
Summarizing what the new technique offers, the research team wrote : "Our proof-of-concept demonstrations point the way to significant potential benefits in these applications by removing the need to" get the job done. " a potentially laborious collection of own data. " Importantly, we must remember the warning" CSI vs reality "; Of course, there is no free meal – we can not learn to grasp the characteristics which are not present in the input data – but this also applies to the training with own targets. "
[19659004] The presentation of the research team will take place on Thursday at the CIML.In the meantime, you can read about the research in more depth by downloading the document (PDF).
Source link
