Panoptic Segmentation using the example
Posted: Sat Jan 25, 2025 8:59 am
Image captioning describes in words what is happening in images. This method combines computer vision with natural language processing. An encoder-decoder framework is usually used, in which an input image is transformed into an intermediate representation that describes the information in the image and then decoded into a descriptive text.
a zebra standing in a field of tall grass
zebra
Denoising / Noise Reduction
Denoising removes noise from an image, i.e. disruptive factors such as incorrect pixel colors. These procedures are important in medicine, as noise is often present in radiological images. On the left side of the image greece consumer email list you can see the patient's lungs with noise, on the right side you can see the denoised version (without/reduced noise).
Medical image of a lung
Super Resolution
Super Resolution is a process that artificially improves the resolution of images. The image on the left is the low-resolution version of the image. The image on the right is the version improved by neural networks. The image has also been enlarged. It is therefore larger, sharper and details are easier to see.
This technology is used to display compressed images in better resolution. The best-known example of this is NVidia DLSS. The resolution of images is improved via such networks, which means that less storage space needs to be used.
Super Resolution as an example
Other applications
There are many other applications of computer vision:
Edge Detection: Detection of edges in images
Surface Normals: Predicting the surface orientation of existing objects
Reshading: Shading refers to the representation of depth perception in 3D models
Uncertainty Estimation: Calculating how inaccurate a prediction is
Depth Estimation: Predicting the depth of objects in an image
Advantages of Deep Learning over Traditional Methods
Traditional methods require domain expertise to explain classes. These descriptive properties are called the descriptive patches of an image.
The next step is to use techniques such as SIFT or BRIEF to describe these features. Patches are detected using edge detection, corner detection and threshold segmentation. As many features as possible are extracted and used as the definition of this class. Then these patches are searched for in other images and if they match, they are assigned to the class.
a zebra standing in a field of tall grass
zebra
Denoising / Noise Reduction
Denoising removes noise from an image, i.e. disruptive factors such as incorrect pixel colors. These procedures are important in medicine, as noise is often present in radiological images. On the left side of the image greece consumer email list you can see the patient's lungs with noise, on the right side you can see the denoised version (without/reduced noise).
Medical image of a lung
Super Resolution
Super Resolution is a process that artificially improves the resolution of images. The image on the left is the low-resolution version of the image. The image on the right is the version improved by neural networks. The image has also been enlarged. It is therefore larger, sharper and details are easier to see.
This technology is used to display compressed images in better resolution. The best-known example of this is NVidia DLSS. The resolution of images is improved via such networks, which means that less storage space needs to be used.
Super Resolution as an example
Other applications
There are many other applications of computer vision:
Edge Detection: Detection of edges in images
Surface Normals: Predicting the surface orientation of existing objects
Reshading: Shading refers to the representation of depth perception in 3D models
Uncertainty Estimation: Calculating how inaccurate a prediction is
Depth Estimation: Predicting the depth of objects in an image
Advantages of Deep Learning over Traditional Methods
Traditional methods require domain expertise to explain classes. These descriptive properties are called the descriptive patches of an image.
The next step is to use techniques such as SIFT or BRIEF to describe these features. Patches are detected using edge detection, corner detection and threshold segmentation. As many features as possible are extracted and used as the definition of this class. Then these patches are searched for in other images and if they match, they are assigned to the class.