Anisotropic Kuwahara Filter #68
Conversation
- New StructuredTensor mode in the EdgeDetect node to output the StructuredTensor - New LineIntegralConvolution node to compute Line Integral Convolution (imaging vector field using flow lines) - New GaussianKernel node that compute and output a ComputeBuffer based on a kernel radius and sigma - New Edge Tangent Flow node that compute the Vector Field (TFM) - New Anisotropic Kuwahara Filter Node that process the painterly filter - New CompareTextures node (only shader, need a Shader Node with the shader assign) TODO : - Add Kuwahara Parameters as Input - Cleanup code - Add Heightmap Generation and composition with Kuwahara Output
felixcantet
changed the title
There was a problem hiding this comment.
What's the license of this image?
| public override Texture previewTexture => output; | ||
| public override bool showDefaultInspector => true; | ||
| [Input("Source")] public Texture source; | ||
| [Input("TFM")] public Texture tfm; |
There was a problem hiding this comment.
Try to avoid using abbreviations in node parameters :)
| if (!base.ProcessNode(cmd) || tfm == null || source == null) | ||
| return false; | ||
| ValidateTempRenderTexture(); | ||
| for (int i = 0; i < passCount; i++) |
There was a problem hiding this comment.
What is an acceptable range of value for the pass count? It's to know if it's worth avoiding the copy inside the loop or not.
| namespace Mixture | ||
| { | ||
| [System.Serializable, NodeMenuItem("Operator/Edge Tengent Flow")] | ||
| public class EdgeTengentFlow : FixedShaderNode |
There was a problem hiding this comment.
Can you write a bit of documentation on this node using the [Documentation] attribute?
| namespace Mixture | ||
| { | ||
| [System.Serializable][NodeMenuItem("Operators/Anisotropic Kuwahara Filter")] | ||
| public class AnisotropicKuwaharaFilter : ComputeShaderNode |
There was a problem hiding this comment.
Can you write a bit of documentation on this node using the [Documentation] attribute?
| { | ||
| var kernel = OneDimensinalKernel(gaussianRadius, gaussianSigma);// GenerateGaussianKernel(gaussianRadius, gaussianSigma); | ||
| var buffer = new ComputeBuffer(kernel.Length, sizeof(float), ComputeBufferType.Default); | ||
| buffer.SetData(kernel); |
There was a problem hiding this comment.
Use CommandBuffer.SetBufferData instead otherwise the execution of the graph and the data will be out of sync.
Alternatively, you can also dispatch a compute that updates this buffer.
| float2 v = mul(SR, float2(x, y)); | ||
| if (dot(v, v) <= 0.25f) | ||
| { | ||
| float3 c = _Source[id.xy + int2(x, y)].rgb; |
There was a problem hiding this comment.
Is there a particular reason to not use a sampler for reading source values?
| output += float4(m[k].rgb * w, w); | ||
| } | ||
|
|
||
| _Output[id.xy] = saturate(output / output.w); |
There was a problem hiding this comment.
Saturate clamps all values outside of 0;1 range, which implicitly means that this algorithm doesn't work outside this range (HDR). Is that correct?
| half4 sumWeight = 0.0h; | ||
|
|
||
| for (int xi = 1; xi <= __KernelRadius; xi++){ | ||
| half2 G = __TFM.SampleLevel(point_clamp_sampler, position, 0).xy; |
There was a problem hiding this comment.
For point sampling Load() is faster, unless you're using the clamp functionality of the sampler in case position goes outside of the [0; 1] bounds
There was a problem hiding this comment.
Where is this shader used?
I think this is something that already exists with the Difference node
|
Thanks for the review ! I still have some features of the nodes to implement and I will make sure to follow your guidelines before asking for review :) |
TODO :