DenoisEM

Frequently asked questions

• Which types of images does DenoisEM support?
  DenoisEM supports 8-bit and 16-bit single channel images. Extending DenoisEM to support multi-channel images is straightforward and planned as future work.


• Can DenoisEM handle images with a large width or height in pixels?
  Yes. DenoisEM will split large images into a set of overlapping tiles. Each tile will be denoised independently to avoid running out of GPU memory. Afterwards the tile overlaps will be discarded and the center portions of the tiles recombined into a seamless result image. This approach completely avoids tile edge artefacts.


• Why does ImageJ refuse to perform certain operations and says my image is locked when I am using DenoisEM?
  During denoising preview and during full denoising, DenoisEM locks the input image to avoid that it changes or disappears behind its back. If you want to modify the image, simply close the DenoisEM wizard to unlock the image.


• Can I change the region of interest (ROI) during algorithm selection and parameter optimization?
  Yes. In the DenoisEM wizard, click Back to return to the "Select Image and ROI" step in the DenoisEM wizard. Then change the ROI either by dragging the existing ROI around, or by drawing a new ROI rectangle. Then click Next to return to the algorithm selection and parameter optimization panel. The denoising algorithm and the user chosen parameter values are not modified when changing the ROI. Similarly, the z-slice on which the denoising preview operates can only be changed while in the ROI selection panel.


• How can I tell lateron which parameters were used to denoise an image using DenoisEM?
  DenoisEM stores the denoising algorithm and its parameters as image properties in ImageJ. These key-value pairs can be inspected in ImageJ via "Image > Properties..." These properties are also saved (and loaded) along with the pixel data to file formats such as TIFF.


• Can I perform "in place" denoising directly on the source image?
  No. Currently DenoisEM always creates a new image or image stack with the denoised result.


• Why is there sometimes temporarily a red frame around the denoising preview? What does it mean?
  The red frame indicates that the denoising calculations cannot be performed at interactive speed anymore. This happens for certain algorithm-parameter combinations that are exceptionally computationally expensive. As long as the red frame is visible, user changes to the algorithm parameters are ignored until DenoisEM completed the current denoising calculations, at which point the red frame disappears. This avoids a buildup of denoising work.


• Can I apply the denoising algorithms with parameters values that are outside the range of the parameter sliders in the DenoisEM user interface?
  Yes. While we tried to offer reasonable parameter ranges based on an estimate of the amount of noise present in the image, the user may wish to experiment with more extreme parameter settings. In that case, simply fill in the desired parameter value in the corresponding numeric edit field instead.


• What about Linux and Mac support?
  The Quasar backend, which is responsible for executing the denoising algorithms on the GPU, is already available for Linux. Mac support for Quasar would be possible too (via OpenCL) but is currently not planned. However, to simplify installation and testing we decided to release the first version of DenoisEM on Windows only.


• Does DenoisEM work in a virtualized environment, such as VirtualBox?
  We do not support DenoisEM running in a virtualized environment. The virtualization layer interferes with Quasar's need to directly access the graphics hardware. Limited testing indicates that DenoisEM will likely work, but will fall back to slower CPU parallelism instead of faster computation on the GPU.