Filtering image data using a neural network
A method, computer readable medium, and system are disclosed for performing spatiotemporal filtering. The method includes identifying image data to be rendered, reconstructing the image data to create reconstructed image data, utilizing a filter including a neural network having one or more skip connections and one or more recurrent layers, and returning the reconstructed image data.
1. A method for performing spatiotemporal filtering, comprising:
identifying image data to be rendered, the image data including:
color data, and
one or more auxiliary features that include at least one of one or more view-space shading normals, one or more depth values, or one or more roughness values;
reconstructing the image data to create reconstructed image data, utilizing a filter including a neural network that processes the image data based on the one or more auxiliary features, the neural network having:
one or more skip connections, and
one or more recurrent layers that retain, at least in part, the one or more auxiliary features unchanged; and
returning the reconstructed image data and the unchanged one or more auxiliary features.
2. The method of claim 1 , wherein the image data is created as a result of a predetermined process, the predetermined process including one or more of:
low light photography;
low light video recording;
a Monte Carlo sampling method;
photography that produces RGB and depth information; and
depth-only cameras.
3. The method of claim 1 , wherein the image data has a number of samples per pixel (spp) below a predetermined threshold.
4. The method of claim 1 , wherein the one or more auxiliary features includes the one or more view-space shading normals.
5. The method of claim 1 , wherein the one or more auxiliary features includes the one or more depth values.
6. The method of claim 1 , wherein the one or more auxiliary features includes the one or more roughness values.
7. The method of claim 1 , wherein the neural network includes a plurality of convolution layers.
8. The method of claim 1 , wherein the neural network includes a plurality of encoder layers and corresponding decoder layers, and a first skip connection within the neural network enables a reconstruction process to jump from a first encoder layer directly to a corresponding decoder layer, thereby skipping over one or more encoder and decoder layers within the neural network.
9. The method of claim 1 , wherein the one or more recurrent layers include a recurrent convolutional neural network (RCNN).
10. The method of claim 1 , wherein the one or more recurrent layers link a layer to itself within the neural network.
11. The method of claim 1 , further comprising training the neural network, utilizing a data set including a plurality of input/output pairs.
12. The method of claim 11 , wherein a loss function is utilized during the training of the neural network.
13. The method of claim 1 , wherein the image data includes a first frame of a sequence of frames, and wherein the one or more auxiliary features used during the reconstructing of the first frame is stored as a hidden, recurrent state, which is then accessed during a reconstruction of a second frame of the sequence of frames of the image data within an RCNN, utilizing a feedback loop of the RCNN.
14. A system comprising:
a processor that is configured to:
identify image data to be rendered, the image data including:
color data, and
one or more auxiliary features that include at least one of one or more view-space shading normals, one or more depth values, or one or more roughness values;
reconstruct the image data to create reconstructed image data, utilizing a filter including a neural network that processes the image data based on the one or more auxiliary features, the neural network having:
one or more skip connections, and
one or more recurrent layers that retain, at least in part, the one or more auxiliary features unchanged; and
return the reconstructed image data and the unchanged one or more auxiliary features.
15. The system of claim 14 , wherein the image data is created as a result of a predetermined process, the predetermined process including one or more of:
low light photography;
low light video recording;
a Monte Carlo sampling method;
photography that produces RGB and depth information; and
depth-only cameras.
16. The system of claim 14 , wherein the image data has a number of samples per pixel (spp) below a predetermined threshold.
17. The system of claim 14 , wherein the one or more auxiliary features includes the one or more view-space shading normals.
18. A method for performing spatiotemporal filtering, comprising:
identifying image data to be rendered, the image data resulting from a Monte Carlo sampling method and the image data including:
color data, and
one or more auxiliary features that include at least one of one or more view-space shading normals, one or more depth values, or one or more roughness values;
applying a filter to the image data to create reconstructed image data, where the filter is a neural network that processes the image data based on the one or more auxiliary features, the neural network having:
one or more skip connections, and
one or more recurrent layers that retain, at least in part, the one or more auxiliary features unchanged; and
returning the reconstructed image data and the unchanged one or more auxiliary features.
19. The method of claim 1 , wherein the image data and the one or more auxiliary features are identified from a deep image consisting of multiple buffers storing the image data and the one or more auxiliary features.
20. The method of claim 11 , wherein the neural network is trained by:
demodulating an input image of an input and output pair to remove texture from the input image;
processing the demodulated input image to reconstruct the demodulated input image; and
generating a final rendering by re-modulating the reconstructed demodulated input image to include the texture.