Jump to content
  • Advertisement
Sign in to follow this  
nbertoa

DX12 [D3D12] SSAO Demo

This topic is 678 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Hi community

 

I want to share a new update of my little DirectX 12 rendering engine. I implemented Screen Space Ambient Occlusion and improved camera movement.

 

Post

 

Video

 

Hope you find it useful!

Share this post


Link to post
Share on other sites
Advertisement
The first video shows the ambient accessibility buffer.

The second video shows the scene where only ambient light is used (ambient factor * diffuse albedo * ambient accessibility)

The third video shows the scene with additional light from irradiance diffuse & specular environment cube maps.

There is no extra light in any scene (i.e. there is no directional light mounted to the camera)

Share this post


Link to post
Share on other sites

I agree with JoeJ, something seems a bit strange with your results. It seems to be view dependent, with the accessibility being higher in places where N dot V is higher. It particularly stands out on the floor, which is darkened despite having any obstructions.

 

Either way I'm not trying to pick on your work, just trying to help in case there's a bug. :)

Edited by MJP

Share this post


Link to post
Share on other sites
@MJP and @Joej, the intention of my posts is to receive feedback about my work (to learn and improve my knowledge). Do not worry! I receive all critics as constructive feedback.

I use normal vector buffer, where normals are stored in view space. Could that affect final result?

Share this post


Link to post
Share on other sites

I guess all AO samples return the result of being visible. In the video the camera rotates around the scene but there never shows up a typical artefact caused by missing SS information.

So it seems the sampling has no effect and the N dot V dependent look might become from elsewhere.

Or depth sampling does not work for some reason? Or all samples at zero distance? Not sure, but i wonder what causes banding if the bug would be something like that.

Share this post


Link to post
Share on other sites

@JoeJ

 

I modified occlusion radius and a lot of problems appeared. I recorded a video that shows them.

 

 

Clearly, it shows that there is view dependent problem (when I move the camera, ambient accessibility changes a lot)

 

This the code that computes ambient occlusion:

float SSAOVersion1(
	const float3 sampleKernel,
	const float3x3 sampleKernelMatrix, 
	const float4x4 projMatrix,
	const float occlusionRadius,
	const float3 fragPosV,
	Texture2D<float> depthTex)
{
	// Get sample position
	float3 sampleV = mul(sampleKernel, sampleKernelMatrix);
	sampleV = sampleV * occlusionRadius + fragPosV;

	// Project sample position
	float4 sampleH = float4(sampleV, 1.0f);
	sampleH = mul(sampleH, projMatrix);
	sampleH.xy /= sampleH.w;
	sampleH.xy = sampleH.xy * 0.5 + 0.5;

	// Get sample depth
	float sampleDepthV = depthTex.Load(float3(sampleH.xy, 0));
	sampleDepthV = NdcDepthToViewDepth(sampleDepthV, projMatrix);

	// Range check and ambient occlusion factor
	const float rangeCheck = abs(fragPosV.z - sampleDepthV) < occlusionRadius ? 1.0 : 0.0;
	return (sampleDepthV <= sampleV.z ? 1.0 : 0.0) * rangeCheck;
}

where sampleKernelMatrix is:

	// Construct a change-of-basis matrix to reorient our sample kernel
	// along the origin's normal.
	const float3 noiseVec = NoiseTexture.Sample(TexSampler, NOISE_SCALE * input.mTexCoordO).xyz * 2.0f - 1.0f ;
	const float3 tangentV = normalize(noiseVec - normalV * dot(noiseVec, normalV));
	const float3 bitangentV = cross(normalV, tangentV);
	const float3x3 sampleKernelMatrix = float3x3(tangentV, bitangentV, normalV);

This is the method that generates sample kernel

	// Sample kernel for ambient occlusion. The requirements are that:
	// - Sample positions fall within the unit hemisphere
	// - Sample positions are more densely clustered towards the origin.
	//   This effectively attenuates the occlusion contribution
	//   according to distance from the kernel centre (samples closer
	//   to a point occlude it more than samples further away).
	void GenerateSampleKernel(const std::uint32_t numSamples, std::vector<XMFLOAT3>& kernels) {
		ASSERT(numSamples > 0U);

		kernels.resize(numSamples);
		XMFLOAT3* data(kernels.data());
		XMVECTOR vec;
		const float numSamplesF = static_cast<float>(numSamples);
		for (std::uint32_t i = 0U; i < numSamples; ++i) {
			XMFLOAT3& elem = data[i];

			// Create sample points on the surface of a hemisphere
			// oriented along the z axis
			const float x = MathUtils::RandF(-1.0f, 1.0f);
			const float y = MathUtils::RandF(-1.0f, 1.0f);
			const float z = MathUtils::RandF(-1.0f, 0.0f);
			elem = XMFLOAT3(x, y, z);
			vec = XMLoadFloat3(&elem);
			vec = XMVector3Normalize(vec);

			// Accelerating interpolation function to falloff 
			// from the distance from the origin.
			float scale = i / numSamplesF;
			scale = MathUtils::Lerp(0.1f, 1.0f, scale * scale);
			vec = XMVectorScale(vec, scale);
			XMStoreFloat3(&elem, vec);
		}
	}

and this is the method that generates noise vectors

	// Generate a set of random values used to rotate the sample kernel,
	// which will effectively increase the sample count and minimize 
	// the 'banding' artifacts.
	void GenerateNoise(const std::uint32_t numSamples, std::vector<XMFLOAT4>& noises) {
		ASSERT(numSamples > 0U);

		noises.resize(numSamples);
		XMFLOAT4* data(noises.data());
		XMVECTOR vec;
		for (std::uint32_t i = 0U; i < numSamples; ++i) {
			XMFLOAT4& elem = data[i];

			// Create sample points on the surface of a hemisphere
			// oriented along the z axis
			const float x = MathUtils::RandF(-1.0f, 1.0f);
			const float y = MathUtils::RandF(-1.0f, 1.0f);
			const float z = 0.0f;			
			elem = XMFLOAT4(x, y, z, 0.0f);
			vec = XMLoadFloat4(&elem);
			vec = XMVector4Normalize(vec);
			XMStoreFloat4(&elem, vec);
			XMFLOAT3 mappedVec = MathUtils::MapF1(XMFLOAT3(elem.x, elem.y, elem.z));
			elem.x = mappedVec.x;
			elem.y = mappedVec.y;
			elem.z = mappedVec.z;
		}
	}
Edited by nicolas.bertoa

Share this post


Link to post
Share on other sites

I attached a screenshot of a wall showing more errors in the ambient accessibility computation.

 

Any ideas?

Share this post


Link to post
Share on other sites

Finally, I fixed the ambient occlusion algorithm. I attached some screenshots with the results.

 

The problem was here

// Project sample position
float4 sampleH = float4(sampleV, 1.0f);
sampleH = mul(sampleH, projMatrix);
sampleH.xy /= sampleH.w;

// Get sample depth
float sampleDepthV = depthTex.Load(float3(sampleH.xy, 0));
sampleDepthV = NdcDepthToViewDepth(sampleDepthV, projMatrix);

because I was using Load() with sampleH that is in NDC space, not in viewport space.

So I did the following

// Convert sample position to NDC and sample depth at that position in depth buffer.
float4 samplePosH = mul(samplePosV, gFrameCBuffer.mP);
samplePosH.xy /= samplePosH.w;
	
const int2 sampleViewportSpace = NdcToViewportCoordinates(samplePosH.xy, 0.0f, 0.0f, SCREEN_WIDTH, SCREEN_HEIGHT);
const float sampleDepthNDC = Depth.Load(int3(sampleViewportSpace, 0));

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

GameDev.net is your game development community. Create an account for your GameDev Portfolio and participate in the largest developer community in the games industry.

Sign me up!