Getting Started with scimesh C++

A gentle introduction to using the scimesh C++ software renderer. No OpenGL, GPU, or display server required — just a C++17 compiler.

What is scimesh C++?

scimesh is a headless software renderer that takes 3D triangle meshes and produces RGBA images entirely on the CPU. It is designed for scientific visualization where GPU access is unavailable or impractical: HPC clusters, CI pipelines, headless servers, and containers.

The renderer supports multi-light Blinn-Phong shading, anti-aliasing, semi-transparent overlays, wireframe mode, SSAO, fog, clip planes, and procedural geometry.

scimesh is not a game engine, raytracer, or interactive viewer. It produces static images.

Prerequisites

• A C++17 compiler (GCC 7+, Clang 5+, MSVC 2017+)
• CMake 3.10 or newer

Project Layout

The C++ renderer core lives in src/core/:

File	Purpose
scimesh/renderer.h	Main Renderer class
scimesh/mesh.h	Mesh struct (vertices, triangles, colors, UVs, normals)
scimesh/scene.h	Scene struct (collection of meshes)
scimesh/camera.h	Camera struct and auto-framing helpers
scimesh/render_options.h	RenderOptions struct (resolution, shading, lights, AA, etc.)
scimesh/image.h	Image struct (RGBA pixel buffer, PPM/BMP output)
scimesh/primitives.h	Procedural geometry generators
scimesh/transforms.h	Mesh translation, scaling, rotation
scimesh/normals.h	Vertex normal computation
scimesh/clipping.h	Triangle clipping against planes
scimesh/rasterizer.h	Scanline rasterizer with z-buffer
scimesh/math_utils.h	Inline math helpers
third_party/glm/	Vendored GLM math library (headers only)

To use scimesh in your own project, compile the .cpp files from src/core/ alongside your code. Headers are under src/core/scimesh/ and included as #include <scimesh/header.h>.

Minimal Example

The simplest program builds a colored cube and renders it:

#include <scimesh/renderer.h>
#include <scimesh/camera.h>
#include <scimesh/render_options.h>
#include <scimesh/primitives.h>
#include <scimesh/image.h>
 
using namespace scimesh;
 
int main() {
    // 1. Create a mesh — a red cuboid
    Mesh cube = generate_cuboid(
        Vec3(0, 0, 0),              // center
        Vec3(1, 1, 1),              // half-extents
        Color(0.9f, 0.2f, 0.2f, 1)  // RGBA color
    );
 
    // 2. Set up camera — auto-frame to fit the mesh
    Vec3 view_dir = glm::normalize(Vec3(1.0f, 1.0f, 1.0f));
    Camera cam = camera_fit_mesh(cube, view_dir,
        Vec3(0, 1, 0),  // up
        45.0f,           // FOV
        1.1f             // margin
    );
 
    // 3. Configure rendering
    RenderOptions opts;
    opts.width  = 800;
    opts.height = 600;
    opts.background_color = Color(1, 1, 1, 1);  // white background
    opts.shading = ShadingMode::SMOOTH;
 
    // 4. Render
    Renderer renderer;
    Image img = renderer.render_mesh(cube, cam, opts);
 
    // 5. Save
    img.write_ppm("cube.ppm");
    img.write_bmp("cube.bmp");
    img.write_tga("cube.tga");   // uncompressed true-color, no dependencies
 
    return 0;
}

Loading a PLY Mesh

Instead of generating geometry, you can load a mesh from a PLY file:

#include <scimesh/renderer.h>
#include <scimesh/camera.h>
#include <scimesh/render_options.h>
#include <scimesh/ply_io.h>
#include <scimesh/image.h>
 
using namespace scimesh;
 
int main() {
    Mesh mesh = ply_io::read("bunny.ply");
 
    Vec3 view_dir = glm::normalize(Vec3(1.0f, 1.0f, 1.0f));
    Camera cam = camera_fit_mesh(mesh, view_dir, Vec3(0, 1, 0), 45.0f, 1.1f);
 
    RenderOptions opts;
    opts.width  = 800;
    opts.height = 600;
    opts.background_color = Color(0.98f, 0.98f, 0.98f, 1);
    opts.shading = ShadingMode::SMOOTH;
 
    Renderer renderer;
    Image img = renderer.render_mesh(mesh, cam, opts);
 
    img.write_ppm("bunny.ppm");
 
    return 0;
}

Compile with ply_io.cpp added to the source list (see Building below).

Building

CMake with FetchContent (Recommended)

The easiest way to use scimesh in your own CMake project is via FetchContent. Add the following to your CMakeLists.txt:

include(FetchContent)
FetchContent_Declare(
    scimesh
    GIT_REPOSITORY https://github.com/dfsp-spirit/scimesh.git
    GIT_TAG        main          # or a release tag, e.g. v0.2.8
)
FetchContent_MakeAvailable(scimesh)
 
add_executable(my_app main.cpp)
target_link_libraries(my_app PRIVATE scimesh)

That's it — CMake will clone scimesh, build it as a static library, and make all headers and compiled code available to your target. No manual source listing, include paths, or preprocessor defines needed.

For the in-repo examples, see any examples/cpp/*/CMakeLists.txt which use add_subdirectory() for the same effect.

Manually (without CMake)

Copy scimesh src/ so it sits next to your project. Assuming your code is in my_project/, the layout should look like this:

parent_directory/
├── my_project/           ← your application
│   ├── main.cpp
│   └── build/
└── src/                  ← scimesh repo src/ copied here
    ├── core/
    └── third_party/

From inside build/, scimesh sources are reachable at ../../src/. Compile the .cpp files from ../../src/core/ alongside ../main.cpp with -std=c++17.

‍Note: If your code calls write_png(), add -DSCIMESH_STB_WRITE_IMPL.

write_tga(), write_bmp() and write_ppm() need no extra defines — they use scimesh's own writers.

A runnable script that manually compiles one of our demos using g++: examples/cpp/all_primitives/build_manually.sh.

Core Concepts

Mesh

A Mesh holds the geometry and appearance data:

struct Mesh {
    std::vector<Vec3>     vertices;     // vertex positions
    std::vector<Triangle> triangles;    // triangle index triplets
    std::vector<Color>    colors;       // per-vertex RGBA (optional)
    std::vector<Color>    face_colors;  // per-face RGBA (optional)
    std::vector<Vec3>     normals;      // per-vertex normals (optional)
    std::vector<Vec2>     uvs;          // texture coordinates (optional)
    Image                 texture;      // texture image (optional)
    Color                 default_color;
    bool                  has_transparency;
};

A Triangle is three indices into the vertex array:

struct Triangle { uint32_t v0, v1, v2; };

Colors are RGBA floats in [0, 1]:

struct Color { float r, g, b, a; };

Scene

A Scene is a collection of meshes rendered together:

Scene scene;
scene.meshes.push_back(mesh_a);
scene.meshes.push_back(mesh_b);

When meshes have semi-transparent colors (has_transparency = true or alpha < 1), the renderer sorts them back-to-front and blends appropriately.

Camera

A Camera defines the viewpoint:

Camera cam;
cam.eye    = Vec3(0, 0, 5);    // where the camera is
cam.center = Vec3(0, 0, 0);    // what it looks at
cam.up     = Vec3(0, 1, 0);    // up direction
cam.projection = ProjectionType::PERSPECTIVE;
cam.fov_degrees = 45.0f;

Use the auto-framing helpers to avoid manual positioning:

// Fit a single mesh
Camera cam = camera_fit_mesh(mesh, view_dir, up, 45.0f, 1.1f);
 
// Fit an entire scene
Camera cam = camera_fit_scene(scene, view_dir, up, 45.0f, 1.1f);

RenderOptions

Controls image size, shading, lighting, and post-processing:

RenderOptions opts;
opts.width  = 1200;
opts.height = 900;
opts.shading = ShadingMode::SMOOTH;        // or FLAT
opts.backface_culling = true;
opts.background_color = Color(0, 0, 0, 0); // transparent
 
// Lighting
opts.ambient = 0.2f;
opts.contrast = 1.1f;
opts.lights = {key_light, fill_light, rim_light};
opts.specular_color = Color(0.4f, 0.4f, 0.4f);
opts.shininess = 64.0f;
 
// Anti-aliasing
opts.aa_samples = 2;  // 2x2 SSAA
 
// SSAO
opts.ssao_enabled = true;
opts.ssao_radius = 12.0f;
opts.ssao_intensity = 0.5f;
 
// Projection
opts.projection = ProjectionType::ORTHOGRAPHIC;

Renderer

The Renderer is the main entry point:

Renderer renderer;
 
// Single mesh
Image img = renderer.render_mesh(mesh, cam, opts);
 
// Multi-mesh scene
Image img = renderer.render_scene(scene, cam, opts);
 
// Raw triangles (no index buffer)
Image img = renderer.render_triangles_raw(positions, colors, cam, opts);
 
// Points
Image img = renderer.render_points_raw(positions, colors, radius, cam, opts);

Image

Image stores RGBA pixels and can write to several formats:

Image img(width, height);
 
img.write_ppm("output.ppm");  // ASCII PPM
img.write_bmp("output.bmp");  // Windows BMP
 
// Direct pixel access
img.set_pixel(x, y, r, g, b, a);
img.get_pixel(x, y, r, g, b, a);
img.clear(r, g, b, a);
 
// Bilinear texture sampling
Color c = img.sample_bilinear(u, v);
 
// Post-processing
img.apply_contrast(1.1f);  // S-curve contrast stretch
 
// Downsample (e.g., for anti-aliasing)
Image small = img.downsample_box(2);

Mesh I/O

scimesh can read and write several mesh formats:

STL

#include <scimesh/stl_io.h>
 
Mesh mesh = stl_io::read("model.stl");
stl_io::write(mesh, "output.stl");         // binary
stl_io::write(mesh, "output.stl", true);   // ASCII

Wavefront OBJ

#include <scimesh/obj_io.h>
 
Mesh mesh = obj_io::read("model.obj");

The OBJ reader supports vertices, faces, UV coordinates, and normals. For textured meshes, the texture image must be loaded separately.

Stanford PLY

#include <scimesh/ply_io.h>
 
Mesh mesh = ply_io::read("model.ply");

The PLY reader supports ASCII and binary formats, with optional per-vertex colors.

Procedural Geometry

Generate primitives without external files:

#include <scimesh/primitives.h>
 
Mesh sphere = generate_sphere(
    Vec3(0, 0, 0),     // center
    1.0f,               // radius
    32,                 // segments
    Color(0, 0.8f, 1, 1)
);
 
Mesh cylinder = generate_cylinder(
    Vec3(0, 0, 0),     // start
    Vec3(0, 3, 0),     // end
    0.5f,               // radius
    16,                 // segments
    Color(1, 0.5f, 0, 1)
);
 
Mesh cuboid = generate_cuboid(
    Vec3(0, 0, 0),     // center
    Vec3(1, 2, 0.5f),  // half-extents
    Color(0.2f, 0.6f, 1, 1)
);
 
Mesh torus = generate_torus(
    Vec3(0, 0, 0),     // center
    1.5f,               // major radius
    0.4f,               // minor radius
    32, 16,             // major/minor segments
    Color(1, 0.5f, 0, 1)
);
 
Mesh cone = generate_cone(
    Vec3(0, 0, 0),     // base
    Vec3(0, 2, 0),     // tip
    0.8f,               // radius
    16,                 // segments
    Color(0.5f, 0, 0.5f, 1)
);
 
Mesh plane = generate_plane(
    Vec3(0, 0, 0),     // center
    Vec3(0, 1, 0),     // normal
    2.0f, 2.0f,         // half-sizes
    Color(0.8f, 0.8f, 0.8f, 1)
);

Merged Primitives

For rendering many instances at once (e.g., molecular atoms), use the merged variants that combine multiple primitives into a single mesh:

#include <scimesh/primitives.h>
 
std::vector<Vec3> centers = {Vec3(0,0,0), Vec3(2,0,0), Vec3(1,2,0)};
std::vector<float> radii = {0.5f, 0.3f, 0.4f};
std::vector<Color> colors = {
    Color(1,0,0,1), Color(0,1,0,1), Color(0,0,1,1)
};
 
Mesh atoms = generate_multi_spheres(centers, radii, colors, 16);
 
// Cylinders between point pairs
Mesh bonds = generate_multi_cylinders(starts, ends, radii, colors, 12);

Mesh Transforms

Transform vertices without modifying colors or normals:

#include <scimesh/transforms.h>
 
translate_mesh(mesh, Vec3(5, 0, 0));
scale_mesh(mesh, 2.0f);
rotate_mesh(mesh, 3.14159f / 4, Vec3(0, 0, 1));  // 45 degrees around Z
 
// Arbitrary 4x4 matrix
glm::mat4 M = glm::translate(glm::mat4(1.0f), glm::vec3(0, 0, 0));
transform_mesh(mesh, M);

Lighting

Scimesh works in display-referred (sRGB) colour space for simplicity. Lights are defined as structs with position, color, and intensity:

Light key_light;
key_light.position  = Vec3(0.5f, 1.0f, 1.0f);
key_light.color     = Color(1.0f, 0.97f, 0.90f);
key_light.intensity = 1.5f;
 
Light fill_light;
fill_light.position  = Vec3(-1.0f, 0.2f, 0.5f);
fill_light.color     = Color(0.4f, 0.5f, 0.8f);
fill_light.intensity = 0.5f;
 
RenderOptions opts;
opts.lights = {key_light, fill_light};
opts.ambient = 0.2f;
opts.gamma = 2.2f;
opts.specular_color = Color(0.4f, 0.4f, 0.4f);
opts.shininess = 64.0f;

When no lights are specified, a single headlight at (0, 0, 1) is used.

Contrast

opts.contrast (default 1.0, no change) applies an S-curve contrast stretch after shading via (value - 0.5) * contrast + 0.5. Values > 1.0 push darks toward black and lights toward white. Typical values: 1.1–1.3 for subtle contrast, up to 1.5 for dramatic.

Ambient

opts.ambient (default 0.3) controls how much light reaches surfaces facing away from the light source. Lower values produce deeper shadows. Typical values are 0.1–0.3.

Post-Processing

Image::apply_contrast() applies the same S-curve to an already rendered image:

Image img = renderer.render_mesh(mesh, cam, opts);
img.apply_contrast(1.1f);  // S-curve contrast stretch
img.write_png("output.png");

Semi-Transparent Meshes

For transparency, set per-vertex colors with alpha < 1 and mark the mesh:

Mesh glass_mesh = /* ... */;
for (auto &c : glass_mesh.colors) c.a = 0.3f;
glass_mesh.has_transparency = true;
 
Scene scene;
scene.meshes.push_back(opaque_mesh);
scene.meshes.push_back(glass_mesh);
 
// The renderer sorts transparent meshes back-to-front
Image img = renderer.render_scene(scene, cam, opts);

Camera Helpers

Auto-framing

// Look from a direction, auto-distance to fit
Camera cam = camera_fit_mesh(mesh,
    Vec3(-1, 0, 0.2f),  // view direction
    Vec3(0, 0, 1),       // up
    45.0f,                // FOV
    1.05f                 // 5% margin
);
 
// Fit an entire scene
Camera cam = camera_fit_scene(scene, view_dir, up, 45.0f, 1.1f);

Orthographic projection

cam.projection = ProjectionType::ORTHOGRAPHIC;
cam.fov_degrees = 40.0f;  // orthographic extent

Orbit paths

camera_orbit() rotates a camera's eye and up around its center:

Camera cam = camera_fit_mesh(mesh, view_dir, up, 45.0f);
for (int i = 0; i < 48; i++) {
    Camera orbit_cam = camera_orbit(cam, Vec3(0, 0, 1), 360.0f / 48 * i);
    Image img = renderer.render_mesh(mesh, orbit_cam, opts);
    // write frame...
}

This is useful for turntable video sequences. See examples/cpp/brain_video/ for a complete example.

Anti-Aliasing

Set aa_samples to 2 or 4 for supersampled anti-aliasing:

opts.aa_samples = 2;  // 2x2 SSAA — renders at 2x resolution, downsamples

Higher values produce smoother edges but use more memory and time.

Examples

The examples/cpp/ directory contains complete, runnable programs:

Example	What it demonstrates
spot_cow/	Textured OBJ mesh with multi-light setup and SSAO
transparency/	Semi-transparent overlays with FreeSurfer surfaces
protein_data_bank_pdb_file/	Protein visualization from PDB files
whole_brain_sulc/	Whole-brain sulcal depth rendering
whole_brain_sulc_fsaverage/	Same on fsaverage template
whole_brain_annot/	Cortical parcellation coloring
brain_video/	Turntable animation — 48 orbit frames via camera_orbit()

Each example has its own CMakeLists.txt. To build and run:

cd examples/cpp/spot_cow
mkdir -p build && cd build
cmake ..
make
./spot_cow

Run all examples at once:

./examples/cpp/run_all_cpp.sh

Common Pitfalls

Black images / nothing visible:

• Check that near/far planes contain your geometry. The defaults work for unit-scale meshes; for very large or distant scenes, adjust opts.near_plane and opts.far_plane.
• Make sure your camera is looking at the mesh. Use camera_fit_mesh() to auto-frame.

Inverted faces / missing triangles:

• Triangle winding order matters for backface culling. If faces disappear when rotating, try opts.backface_culling = false or opts.invert_normals = true.

Open surfaces:

• For surface that are open (not watertight), set opts.backface_culling = false so both sides are visible.

Transparency not working:

• Make sure the mesh has has_transparency = true and vertex colors with alpha < 1. Transparent meshes are rendered back-to-front.
• Note that you can also set the background color for the rendered image to transparent.

Performance:

• A 300k triangle mesh at 1200x900 with 2x AA renders in ~1-3 seconds. Higher AA samples and SSAO add cost. Use opts.threads = 0 for auto-parallelization (requires OpenMP).