An easy explanation to Dual Contouring?

Recommended Posts

Hello,

I'm a trainee for software development and in my free time I try to do various things. I decided I wanted to figure out how "Dual Contouring" works, but I haven't been successful yet. I could find some implementations in code which are very hard to follow along and same sheets of paper "explaining" it written in a way that I have a hard time even understanding what the input and output is. All I know that it is used to make  a surface/mesh out of voxels. Is there any explanation someone can understand without having studied math/computer science? The problem is also that most of the words I can't even translate to German(like Hermite Data) nor I can find a explanation which I can understand. I know how Marching Cubes work but I just can't find anything to Dual Contouring also I don't quite get the sense of octrees. As far I'm aware of this is a set of data organized like a tree where each element points to 8 more elements. I don't get how that could be helpful in a infinite 3D world and why I shouldn't just use a List of coordinates with zeros and ones for dirt/air or something like that.

Thanks

A clueless trainee ^^

Edited by QuesterDesura

Share on other sites

Have you seen: https://github.com/emilk/Dual-Contouring ?

It compiles and runs just fine for me on MacOS X. No such luck compiling when using Visual Studio (cl).

I too have been looking forward to familiarizing myself with Dual Contouring.

Share on other sites

I have looked through the code, but I have some issues with the variable names and functions used. As far what I have seen the Field consists of a array of planes, what is kinda confusing as I thought Dual Contouring is only using like a 3D array of zeros and ones. Also I'm not quite sure if a voxel is in the middle of a cube or on the corner because it seems like it is iterating with the foreach3D(function?) through each voxel and determines if the surrounding voxels are all outside or inside(below zero and greater zero as it seems). If thats not the case then it makes vertexes to CrossingCorners as it seems. Is that correct for the first part? I find this confusing because voxels in Minecraft terms are simply displaying 1 Cube and can either be 0 or a ID of a block, but here it seems  to be that 8 voxels/coordinates represent 1 Cube or whatever might come out of the algorithm.

Edited by QuesterDesura

Share on other sites

Have you tried emailing the author of that code? emil.ernerfeldt@gmail.com

Share on other sites
39 minutes ago, sjhalayka said:

Have you tried emailing the author of that code? emil.ernerfeldt@gmail.com

Just did so, I hope he is willing to help me ^^

Share on other sites

Cool. Let me know how it goes.

Share on other sites

Hermite data: it's the same volume data (or distance field) that you'd use for regular marching cubes, but with surface normals at each sampled point as well.

If you are using a distance field as your source of volume data, you can find the normal at each sample point using a central difference, and voila, you have hermite data.

Share on other sites

Conceptually, Dual Contouring is not that different from Marching Cubes. In both cases, you divide your volume data into a grid, and at each cube in the grid you evaluate the surface. Any cube containing edge-crossing - or in other words, any cube that intersects the surface (i.e. isn't entirely empty or entirely full) will generate output geometry.

The key difference between the two is that in Marching Cubes you will just directly generate geometry from the edge crossings. In Dual Contouring you will instead use the normals at each edge-crossing (this is why you need Hermite Data) to locate a single exact point on the surface, and you'll generate geometry from these points (spanning multiple cubes) in a subsequent pass.

The complicated part of Dual Contouring is mostly in how you find the optimal point based on the normals of neighbouring edge crossings. The original paper uses QEF minimisation, which is... a little opaque to those of us without a deep math background.

If you are looking for a more readable implementation, Lin has a pretty simple one on github.

Share on other sites
25 minutes ago, swiftcoder said:

Conceptually, Dual Contouring is not that different from Marching Cubes. In both cases, you divide your volume data into a grid, and at each cube in the grid you evaluate the surface. Any cube containing edge-crossing - or in other words, any cube that intersects the surface (i.e. isn't entirely empty or entirely full) will generate output geometry.

The key difference between the two is that in Marching Cubes you will just directly generate geometry from the edge crossings. In Dual Contouring you will instead use the normals at each edge-crossing (this is why you need Hermite Data) to locate a single exact point on the surface, and you'll generate geometry from these points (spanning multiple cubes) in a subsequent pass.

The complicated part of Dual Contouring is mostly in how you find the optimal point based on the normals of neighbouring edge crossings. The original paper uses QEF minimisation, which is... a little opaque to those of us without a deep math background.

If you are looking for a more readable implementation, Lin has a pretty simple one on github.

Wow, thank you so much you just cleared up so much stuff I found to be confusing. I don't quite understand how those normals should be working. If for example in my game a Cube gets placed by the player then I will create volume data containing the coordinates of the 8 edges, right? But what normal direction does I give the 8 coordinates and how exactly do I use it? Also how should I locate a point on the surface? I thought Dual Contouring is creating this surface out of a set of data, so how can I determine a location on a shape then?

I made a image on how I thought it is working:

1. User wants to create a cube

2. create 8 voxels

4.Dual Contouring

Is this how it works or did I miss something? And how does Dual Contouring actually use those normals?

Edited by QuesterDesura

Share on other sites

I took a shot at deciphering the emilk code, and I'm sort of half-triumphant.

I put up a copy of the code in my own GitHub repository:

I added a function in main.cpp called addFractal (also see below for source code), which adds a quaternion Julia set. I set the quaternion parameters to all zeroes, which produces a 3D ball. If you take a look at the mesh.obj that I put up (also see attached file below), you can make out part of the sphere, but the surface is all choppy. I'm not sure if I'm setting the dist variable or the normal variable correctly. Anyone have any suggestions?

Attached file:

void addFractal(Field& field)
{
float x_grid_max = 1.5;
float y_grid_max = 1.5;
float z_grid_max = 1.5;
float x_grid_min = -x_grid_max;
float y_grid_min = -y_grid_max;
float z_grid_min = -z_grid_max;

float z_w = 0;
quaternion C;
C.x = 0.0f;
C.y = 0.0f;
C.z = 0.0f;
C.w = 0.0f;
unsigned short int max_iterations = 8;
float threshold = 4;

string error_string;
quaternion_julia_set_equation_parser eqparser;
if (false == eqparser.setup("Z = sin(Z) + C * sin(Z)", error_string, C))
{
cout << "Equation error: " << error_string << endl;
return;
}
else
{
cout << "Equation ok" << endl;
}

quaternion Z(x_grid_min, y_grid_min, z_grid_min, z_w);

foreach3D(field.size(), [&](Vec3u upos)
{
auto& cell = field[upos];

Z.x = upos.x / float(FieldSize - 1);
Z.y = upos.y / float(FieldSize - 1);
Z.z = upos.z / float(FieldSize - 1);

cell.dist = eqparser.iterate(Z, max_iterations, threshold) - threshold;

Z.x -= 0.01;
float normal_x1 = eqparser.iterate(Z, max_iterations, threshold);

Z.x += 0.02;
float normal_x2 = eqparser.iterate(Z, max_iterations, threshold);

Z.x -= 0.01;
float normal_x = normal_x1 - normal_x2;

Z.y -= 0.01;
float normal_y1 = eqparser.iterate(Z, max_iterations, threshold);

Z.y += 0.02;
float normal_y2 = eqparser.iterate(Z, max_iterations, threshold);

Z.y -= 0.01;
float normal_y = normal_y1 - normal_y2;

Z.z -= 0.01;
float normal_z1 = eqparser.iterate(Z, max_iterations, threshold);

Z.z += 0.02;
float normal_z2 = eqparser.iterate(Z, max_iterations, threshold);

Z.z -= 0.01;
float normal_z = normal_z1 - normal_z2;

normal_x /= 2.0;
normal_y /= 2.0;
normal_z /= 2.0;

cell.normal = normalized(Vec3(normal_x, normal_y, normal_z));
});
}

Edited by sjhalayka

Create an account

Register a new account

• Similar Content

• By nilkun
Hello everyone!
First time posting in the forum.
I've just completed my first game ever ( C++ / SDL ), and I am feeling utterly proud. It is a small game resembling Missile Command. The code is a mess, but it is my mess! In the process of making the game, I developed my own little game engine.
My question is, where would be a good place to spread the news to at least get some people to try the game?
• By owenjr
Hi, I'm a Multimedia Engineering student. I am about to finish my dergree and I'm already thinking about what topic to cover in my final college project.
I'm interested in the procedural animation with c++ and OpenGL of creatures, something like a spider for example. Can someone tell me what are the issues I should investigate to carry it out? I understand that it has some dependence on artificial intelligence but I do not know to what extent. Can someone help me to find information about it? Thank you very much.

Examples:
- Procedural multi-legged walking animation
- Procedural Locomotion of Multi-Legged Characters in Dynamic Environments

• I am a talented 2D/3D artist with 3 years animation working experience and a Degree in Illustration and Animation. I have won a world-wide art competition hosted by SFX magazine and am looking to develop a survival game. I have some knowledge of C sharp and have notes for a survival based game with flexible storyline and PVP. Looking for developers to team up with. I can create models, animations and artwork and I have beginner knowledge of C sharp with Unity. The idea is Inventory menu based gameplay and is inspired by games like DAYZ.
Here is some early sci-fi concept art to give you an idea of the work level. Hope to work with like minded people and create something special. email me andrewparkesanim@gmail.com.
Developers who share the same passion please contact me, or if you have a similar project and want me to join your team email me.
Many thanks, Andrew.

• By mike44
Hi
saw in dependency walker that my app still needs msvcp140d.dll even after disabling debug.
What did I forget in the VS2017 release settings? After setting to multithreaded dll I get linker errors.
Thanks

• So I have been playing around with yaml-cpp as I want to use YAML for most of my game data files however I am running into some pretty big performance issues and not sure if it is something I am doing or the library itself.
I created this code in order to test a moderately sized file:
Player newPlayer = Player(); newPlayer.name = "new player"; newPlayer.maximumHealth = 1000; newPlayer.currentHealth = 1; Inventory newInventory; newInventory.maximumWeight = 10.9f; for (int z = 0; z < 10000; z++) { InventoryItem* newItem = new InventoryItem(); newItem->name = "Stone"; newItem->baseValue = 1; newItem->weight = 0.1f; newInventory.items.push_back(newItem); } YAML::Node newSavedGame; newSavedGame["player"] = newPlayer; newSavedGame["inventory"] = newInventory; This is where I ran into my first issue, memory consumption.
Before I added this code, the memory usage of my game was about 22MB. After I added everything expect the YAML::Node stuff, it went up to 23MB, so far nothing unexpected. Then when I added the YAML::Node and added data to it, the memory went up to 108MB. I am not sure why when I add the class instance it only adds like 1MB of memory but then copying that data to a YAML:Node instance, it take another 85MB of memory.
So putting that issue aside, I want want to test the performance of writing out the files. the initial attempt looked like this:
void YamlUtility::saveAsFile(YAML::Node node, std::string filePath) { std::ofstream myfile; myfile.open(filePath); myfile << node << std::endl; myfile.close(); } To write out the file (that ends up to be about 570KB), it took about 8 seconds to do that. That seems really slow to me.
After read the documentation a little more I decide to try a different route using the YAML::Emitter, the implemntation looked like this:
static void buildYamlManually(std::ofstream& file, YAML::Node node) { YAML::Emitter out; out << YAML::BeginMap << YAML::Key << "player" << YAML::Value << YAML::BeginMap << YAML::Key << "name" << YAML::Value << node["player"]["name"].as<std::string>() << YAML::Key << "maximumHealth" << YAML::Value << node["player"]["maximumHealth"].as<int>() << YAML::Key << "currentHealth" << YAML::Value << node["player"]["currentHealth"].as<int>() << YAML::EndMap; out << YAML::BeginSeq; std::vector<InventoryItem*> items = node["inventory"]["items"].as<std::vector<InventoryItem*>>(); for (InventoryItem* const value : items) { out << YAML::BeginMap << YAML::Key << "name" << YAML::Value << value->name << YAML::Key << "baseValue" << YAML::Value << value->baseValue << YAML::Key << "weight" << YAML::Value << value->weight << YAML::EndMap; } out << YAML::EndSeq; out << YAML::EndMap; file << out.c_str() << std::endl; } While this did seem to improve the speed, it was still take about 7 seconds instead of 8 seconds.
Since it has been a while since I used C++ and was not sure if this was normal, I decided to for testing just write a simple method to manually generate the YAMLin this use case, that looked something like this:
static void buildYamlManually(std::ofstream& file, SavedGame savedGame) { file << "player: \n" << " name: " << savedGame.player.name << "\n maximumHealth: " << savedGame.player.maximumHealth << "\n currentHealth: " << savedGame.player.currentHealth << "\ninventory:" << "\n maximumWeight: " << savedGame.inventory.maximumWeight << "\n items:"; for (InventoryItem* const value : savedGame.inventory.items) { file << "\n - name: " << value->name << "\n baseValue: " << value->baseValue << "\n weight: " << value->weight; } } This wrote the same file and it took about 0.15 seconds which seemed a lot more to what I was expecting.
While I would expect some overhead in using yaml-cpp to manage and write out YAML files, it consuming 70X+ the amount of memory and it being 40X+ slower in writing files seems really bad.
I am not sure if I am doing something wrong with how I am using yaml-cpp that would be causing this issue or maybe it was never design to handle large files but was just wondering if anyone has any insight on what might be happening here (or an alternative to dealing with YAMLin C++)?

• So I am trying to using Yaml as my game data files (mainly because it support comments, is a bit easier to read than JSON, and I am going to be working in these files a lot) with C++ and yaml-cpp (https://github.com/jbeder/yaml-cpp) seems like the most popular library for dealing with it however I am running into an issue when using pointers.
Here is my code:
struct InventoryItem { std::string name; int baseValue; float weight; }; struct Inventory { float maximumWeight; std::vector<InventoryItem*> items; }; namespace YAML { template <> struct convert<InventoryItem*> { static Node encode(const InventoryItem* inventoryItem) { Node node; node["name"] = inventoryItem->name; node["baseValue"] = inventoryItem->baseValue; node["weight"] = inventoryItem->weight; return node; } static bool decode(const Node& node, InventoryItem* inventoryItem) { // @todo validation inventoryItem->name = node["name"].as<std::string>(); inventoryItem->baseValue = node["baseValue"].as<int>(); inventoryItem->weight = node["weight"].as<float>(); return true; } }; template <> struct convert<Inventory> { static Node encode(const Inventory& inventory) { Node node; node["maximumWeight"] = inventory.maximumWeight; node["items"] = inventory.items; return node; } static bool decode(const Node& node, Inventory& inventory) { // @todo validation inventory.maximumWeight = node["maximumWeight"].as<float>(); inventory.items = node["items"].as<std::vector<InventoryItem*>>(); return true; } }; } if I just did std::vector<InventoryItem> items and had the encode / decode use InventoryItem& inventoryItem everything works fine however when I use the code above that has it as a pointer, I get the following error from code that is part of the yaml-cpp library:
impl.h(123): error C4700: uninitialized local variable 't' used The code with the error is:
template <typename T> struct as_if<T, void> { explicit as_if(const Node& node_) : node(node_) {} const Node& node; T operator()() const { if (!node.m_pNode) throw TypedBadConversion<T>(node.Mark()); T t; if (convert<T>::decode(node, t)) // NOTE: THIS IS THE LINE THE COMPILER ERROR IS REFERENCING return t; throw TypedBadConversion<T>(node.Mark()); } }; With my relative lack of experience in C++ and not being able to find any documentation for yaml-cpp using pointers, I am not exactly sure what is wrong with my code.
Anyone have any ideas what I need to change with my code?

• I already asked this question on stack overflow, and they got pissed at me, down-voted me and so forth, LOL .... so I'm pretty sure the answer is NO, but I'll try again here anyway just in case..... Is there any way to get the size of a polymorphic object at run-time? I know you can create a virtual function that returns size and overload it for each child class, but I'm trying to avoid that since (a) it takes a virtual function call and I want it to be fast and (b) it's a pain to have to include the size function for every subclass. I figure since each object has a v-table their should be some way since the information is there, but perhaps there is no portable way to do it.

• This is the code I have:
//Create Window     DWORD windowStyle = WS_VISIBLE;     DWORD windowExStyle = WS_EX_OVERLAPPEDWINDOW;     SetThreadDpiAwarenessContext(DPI_AWARENESS_CONTEXT_SYSTEM_AWARE);     RECT client{ 0, 0, 100, 40 };     UINT dpi = GetDpiForSystem();     AdjustWindowRectExForDpi(&client, windowStyle, false, windowExStyle, dpi);     UINT adjustedWidth = client.right - client.left;     UINT adjustedHeight = client.bottom - client.top;     m_hwnd = CreateWindowEx(windowExStyle,                             className.c_str(),                             windowName.c_str(),                             windowStyle,                             CW_USEDEFAULT,                             CW_USEDEFAULT,                             adjustedWidth,                             adjustedHeight,                             nullptr,                             nullptr,                             m_hInstance,                             m_emu     ); The generated window has a client area of 1 pixel in height, even though I'm asking for 40. so I'm always getting 39 pixel less than what I need...can someone help me with this? x_x

• I've spent quite a while (and probably far longer than I actually should) trying to design an allocator system.  I've bounced ideas around to various people in the past, but never really gotten something satisfactory.
Basically, the requirements I'm trying to target are:
Composability -- allocators that seamlessly allocate from memory allocated by other allocators.  This helps me to do things like, for example, write an allocator that pads allocations from its parent allocator with bit patterns to detect heap corruption.  It also allows me to easily create spillovers, or optionally assert on overflow with specialized fallbacks.   Handling the fact that some allocators have different interfaces than others in an elegant way.  For example, a regular allocator might have Allocate/Deallocate, but a linear allocator can't do itemized deallocation (but can deallocate everything at once).   I want to be able to tell how much I've allocated, and how much of that is actually being used.  I also want to be able to bucket that on subsystem, but as far as I can tell, that doesn't really impact the design outside of adding a new parameter to allocate calls. Note:  I'm avoiding implementation of allocation buckets and alignment from this, since it's largely orthogonal to what I'm asking and can be done with any of the designs.

To meet those three requirements, I've come up with the following solutions, all of which have significant drawbacks.
Static Policy-Based Allocators
I originally built this off of this talk.
Examples;
struct AllocBlock { std::byte* ptr; size_t size; }; class Mallocator { size_t allocatedMemory; public: Mallocator(); AllocBlock Allocate(size_t size); void Deallocate(AllocBlock blk); }; template <typename BackingAllocator, size_t allocSize> class LinearAllocator : BackingAllocator { AllocBlock baseMemory; char* ptr; char* end; public: LinearAllocator() : baseMemory(BackingAllocator::Allocate(allocSize)) { /* stuff */ } AllocBlock Allocate(size_t size); }; template <typename BackingAllocator, size_t allocSize> class PoolAllocator : BackingAllocator { AllocBlock baseMemory; char* currentHead; public: PoolAllocator() : baseMemory(BackingAllocator::Allocate(allocSize)) { /* stuff */ } void* Allocate(); // note the different signature. void Deallocate(void*); }; // ex: auto allocator = PoolAllocator<Mallocator, size>; Advantages:
SFINAE gives me a pseudo-duck-typing thing.  I don't need any kind of common interfaces, and I'll get a compile-time error if I try to do something like create a LinearAllocator backed by a PoolAllocator. It's composable. Disadvantages:
Composability is type composability, meaning every allocator I create has an independent chain of compositions.  This makes tracking memory usage pretty hard, and presumably can cause me external fragmentation issues.  I might able to get around this with some kind of singleton kung-fu, but I'm unsure as I don't really have any experience with them. Owing to the above, all of my customization points have to be template parameters because the concept relies on empty constructors.  This isn't a huge issue, but it makes defining allocators cumbersome. Dynamic Allocator Dependency
This is probably just the strategy pattern, but then again everything involving polymorphic type composition looks like the strategy pattern to me. 😃
Examples:
struct AllocBlock { std::byte* ptr; size_t size; }; class Allocator { virtual AllocBlock Allocate(size_t) = 0; virtual void Deallocate(AllocBlock) = 0; }; class Mallocator : Allocator { size_t allocatedMemory; public: Mallocator(); AllocBlock Allocate(size_t size); void Deallocate(AllocBlock blk); }; class LinearAllocator { Allocator* backingAllocator; AllocBlock baseMemory; char* ptr; char* end; public: LinearAllocator(Allocator* backingAllocator, size_t allocSize) : backingAllocator(backingAllocator) { baseMemory = backingAllocator->Allocate(allocSize); /* stuff */ } AllocBlock Allocate(size_t size); }; class PoolAllocator { Allocator* backingAllocator; AllocBlock baseMemory; char* currentHead; public: PoolAllocator(Allocator* backingAllocator, size_t allocSize) : backingAllocator(backingAllocator) { baseMemory = backingAllocator->Allocate(allocSize); /* stuff */ } void* Allocate(); // note the different signature. void Deallocate(void*); }; // ex: auto allocator = PoolAllocator(someGlobalMallocator, size); There's an obvious problem with the above:  Namely that PoolAllocator and LinearAllocator don't inherit from the generic Allocator interface.  They can't, because their interfaces provide different semantics.  There's to ways I can solve this:
Inherit from Allocator anyway and assert on unsupported operations (delegates composition failure to runtime errors, which I'd rather avoid).   As above:  Don't inherit and just deal with the fact that some composability is lost (not ideal, because it means you can't do things like back a pool allocator with a linear allocator) As for the overall structure, I think it looks something like this:
Memory usage tracking is easy, since I can use the top-level mallocator(s) to keep track of total memory allocated, and all of the leaf allocators to track of used memory.  How to do that in particular is outside the scope of what I'm asking about, but I've got some ideas. I still have composability Disadvantages:
The interface issues above.  There's no duck-typing-like mechanism to help here, and I'm strongly of the opinion that programmer errors in construction like that should fail at compile-time, not runtime. Composition on Allocated Memory instead of Allocators
This is probably going to be somewhat buggy and poorly thought, since it's just an idea rather than something I've actually tried.
Examples:
struct AllocBlock { void* ptr; size_t size; std::function<void()> dealloc; } class Mallocator { size_t allocatedMemory; public: Mallocator(); AllocBlock Allocate(size_t size) { void* ptr = malloc(size); return {ptr, size, [ptr](){ free(ptr); }}; } }; class LinearAllocator { AllocBlock baseMemory; char* ptr; char* end; public: LinearAllocator(AllocBlock baseMemory) : baseMemory(baseMemory) {end = ptr = baseMemory.ptr;} AllocBlock Allocate(size_t); }; class PoolAllocator { AllocBlock baseMemory; char* head; public: PoolAllocator(AllocBlock baseMemory) : baseMemory(baseMemory) { /* stuff */ } void* Allocate(); }; // ex: auto allocator = PoolAllocator(someGlobalMallocator.Allocate(size)); I don't really like this design at first blush, but I haven't really tried it.

"Composable", since we've delegated most of what composition entails into the memory block rather than the allocator. Tracking memory is a bit more complex, but I *think* it's still doable. Disadvantages:
Makes the interface more complex, since we have to allocate first and then pass that block into our "child" allocator. Can't do specialized deallocation (i.e. stack deallocation) since the memory blocks don't know anything about their parent allocation pool.  I might be able to get around this though.
I've done a lot of research against all of the source-available engines I can find, and it seems like most of them either have very small allocator systems or simply don't try to make them composable at all (CryEngine does this, for example).  That said, it seems like something that should have a lot of good examples, but I can't find a whole lot.  Does anyone have any good feedback/suggestions on this, or is composability in general just a pipe dream?

• Hi
I’ve been working on a game engine for years and I’ve recently come back to it after a couple of years break.  Because my engine uses DirectX9.0c I thought maybe it would be a good idea to upgrade it to DX11. I then installed Windows 10 and starting tinkering around with the engine trying to refamiliarise myself with all the code.
It all seems to work ok in the new OS but there’s something I’ve noticed that has caused a massive slowdown in frame rate. My engine has a relatively sophisticated terrain system which includes the ability to paint roads onto it, ala CryEngine. The roads are spline curves and built up with polygons matching the terrain surface. It used to work perfectly but I’ve noticed that when I’m dynamically adding the roads, which involves moving the spline curve control points around the surface of the terrain, the frame rate comes to a grinding halt.
There’s some relatively complex processing going on each time the mouse moves - the road either side of the control point(s) being moved, is reconstructed in real time so you can position and bend the road precisely. On my previous OS, which was Win2k Pro, this worked really smoothly and in release mode there was barely any slow down in frame rate, but now it’s unusable. As part of the road reconstruction, I lock the vertex and index buffers and refill them with the new values so my question is, on windows 10 using DX9, is anyone aware of any locking issues? I’m aware that there can be contention when locking buffers dynamically but I’m locking with LOCK_DISCARD and this has never been an issue before.
Any help would be greatly appreciated.

• 48
• 12
• 10
• 10
• 9
• Forum Statistics

• Total Topics
631379
• Total Posts
2999666
×