Hi everyone,
I implemented the normal mapping for my shader terrain. Everything works fine for the content in the main window (See the first image below). However, I made a second buffer with a new RGB camera, and the content captured by this new camera is incorrect (See the second image).
My fragment shader is as follows:
#version 330
// Number of splits in the PSSM, it must be in line with what is configured in the PSSMCameraRig
const int split_count=2;
uniform vec3 light_direction;
uniform mat3 p3d_NormalMatrix;
uniform struct {
sampler2D data_texture;
sampler2D heightfield;
int view_index;
int terrain_size;
int chunk_size;
} ShaderTerrainMesh;
uniform struct {
vec4 position;
vec3 color;
vec3 attenuation;
vec3 spotDirection;
float spotCosCutoff;
float spotExponent;
sampler2DShadow shadowMap;
mat4 shadowViewMatrix;
} p3d_LightSource[1];
uniform struct {
vec4 ambient;
} p3d_LightModel;
uniform sampler2D p3d_Texture0;
uniform vec3 wspos_camera;
// asset
uniform sampler2D yellow_tex;
uniform sampler2D white_tex;
uniform sampler2D road_tex;
uniform sampler2D road_normal;
uniform sampler2D road_rough;
uniform sampler2D grass_tex;
uniform sampler2D grass_normal;
uniform sampler2D grass_rough;
uniform float grass_tex_ratio;
uniform sampler2D rock_tex;
uniform sampler2D rock_normal;
uniform sampler2D rock_rough;
uniform sampler2D attribute_tex;
// just learned that uniform means the variable won't change in each stage, while in/out is able to do that : )
uniform float elevation_texture_ratio;
uniform float height_scale;
uniform sampler2D PSSMShadowAtlas;
uniform mat4 pssm_mvps[split_count];
uniform vec2 pssm_nearfar[split_count];
uniform float border_bias;
uniform float fixed_bias;
uniform bool use_pssm;
uniform bool fog;
in vec2 terrain_uv;
in vec3 vtx_pos;
in vec4 projecteds[1];
out vec4 color;
// Projects a point using the given mvp
vec3 project(mat4 mvp, vec3 p) {
vec4 projected = mvp * vec4(p, 1);
return (projected.xyz / projected.w) * vec3(0.5) + vec3(0.5);
}
vec3 get_color(vec3 diffuse, sampler2D normal_tex, sampler2D rough_tex, float tex_ratio, mat3 tbn){
// float roughness = texture(rough_tex, terrain_uv * tex_ratio).r;
vec3 normal = normalize(texture(normal_tex, terrain_uv * tex_ratio).rgb*2.0-1.0) * p3d_NormalMatrix;
// normal = normalize(normal + (roughness - 0.5) * 2.0);
vec3 basecolor = diffuse.xyz;
normal = normalize(tbn * normal);
vec3 light_dir = normalize(light_direction);
vec3 shading = max(0.0, dot(normal, light_dir)) * diffuse;
return shading;
}
void main() {
float road_tex_ratio = 32.0 * elevation_texture_ratio;
float grass_tex_ratio = grass_tex_ratio * elevation_texture_ratio;
float r_min = (1-1/elevation_texture_ratio)/2;
float r_max = (1-1/elevation_texture_ratio)/2+1/elevation_texture_ratio;
vec4 attri = texture(attribute_tex, terrain_uv*elevation_texture_ratio+0.5);
// terrain normal
vec3 pixel_size = vec3(1.0, -1.0, 0) / textureSize(ShaderTerrainMesh.heightfield, 0).xxx;
float h_u0 = texture(ShaderTerrainMesh.heightfield, terrain_uv + pixel_size.yz).x * height_scale;
float h_u1 = texture(ShaderTerrainMesh.heightfield, terrain_uv + pixel_size.xz).x * height_scale;
float h_v0 = texture(ShaderTerrainMesh.heightfield, terrain_uv + pixel_size.zy).x * height_scale;
float h_v1 = texture(ShaderTerrainMesh.heightfield, terrain_uv + pixel_size.zx).x * height_scale;
vec3 tangent = normalize(vec3(1, 0, h_u1 - h_u0));
vec3 binormal = normalize(vec3(0, 1, h_v1 - h_v0));
vec3 terrain_normal = normalize(cross(tangent, binormal));
vec3 normal = normalize(p3d_NormalMatrix * terrain_normal);
// normal.x *= -1;
mat3 tbn = mat3(tangent, binormal, normal);
vec3 shading = vec3(0.0);
// Calculate the shading of each light in the scene
for (int i = 0; i < p3d_LightSource.length(); ++i) {
vec3 diff = p3d_LightSource[i].position.xyz - vtx_pos * p3d_LightSource[i].position.w;
vec3 light_vector = normalize(diff);
vec3 light_shading = clamp(dot(normal, light_vector), 0.0, 1.0) * p3d_LightSource[i].color;
// If PSSM is not used, use the shadowmap from the light
// This is deeply ineficient, it's only to be able to compare the rendered shadows
if (!use_pssm) {
vec4 projected = projecteds[i];
// Apply a bias to remove some of the self-shadow acne
projected.z -= fixed_bias * 0.01 * projected.w;
light_shading *= textureProj(p3d_LightSource[i].shadowMap, projected);
}
shading += light_shading;
}
vec3 color_origin;
if ((attri.r > 0.01) && terrain_uv.x>r_min && terrain_uv.y > r_min && terrain_uv.x<r_max && terrain_uv.y<r_max){
float value = attri.r; // Assuming it's a red channel texture
vec3 diffuse;
if (value < 0.11) {
// Semantics for value 1
diffuse=texture(yellow_tex, terrain_uv * road_tex_ratio).rgb;
} else if (value < 0.21) {
// Semantics for value 2
diffuse = texture(road_tex, terrain_uv * road_tex_ratio).rgb;
} else{
// Semantics for value 4
diffuse = texture(white_tex, terrain_uv * road_tex_ratio).rgb;
}
color_origin=get_color(diffuse, road_normal, road_rough, road_tex_ratio, tbn);
}
else{
// texture splatting, mixing ratio can be determined via rgba, no grass here
vec3 diffuse = texture(grass_tex, terrain_uv * grass_tex_ratio).rgb;
color_origin=get_color(diffuse, grass_normal, grass_rough, grass_tex_ratio, tbn);
}
// static shadow
vec3 light_dir = normalize(light_direction);
shading *= max(0.0, dot(terrain_normal, light_dir));
shading += vec3(0.07, 0.07, 0.1);
// dynamic shadow
if (use_pssm) {
// Find in which split the current point is present.
int split = 99;
float border_bias = 0.5 - (0.5 / (1.0 + border_bias));
// Find the first matching split
for (int i = 0; i < split_count; ++i) {
vec3 coord = project(pssm_mvps[i], vtx_pos);
if (coord.x >= border_bias && coord.x <= 1 - border_bias &&
coord.y >= border_bias && coord.y <= 1 - border_bias &&
coord.z >= 0.0 && coord.z <= 1.0) {
split = i;
break;
}
}
// Compute the shadowing factor
if (split < split_count) {
// Get the MVP for the current split
mat4 mvp = pssm_mvps[split];
// Project the current pixel to the view of the light
vec3 projected = project(mvp, vtx_pos);
vec2 projected_coord = vec2((projected.x + split) / float(split_count), projected.y);
// Apply a fixed bias based on the current split to diminish the shadow acne
float ref_depth = projected.z - fixed_bias * 0.001 * (1 + 1.5 * split);
// Check if the pixel is shadowed or not
float depth_sample = textureLod(PSSMShadowAtlas, projected_coord, 0).x;
float shadow_factor = step(ref_depth, depth_sample);
shading *= shadow_factor;
}
}
shading += p3d_LightModel.ambient.xyz;
shading *= color_origin.xyz;
if (fog) {
// Fake fog
float dist = distance(vtx_pos, wspos_camera);
float fog_factor = smoothstep(0, 1, dist / 8000.0);
shading = mix(shading, vec3(0.7, 0.7, 0.8), fog_factor);
}
color = vec4(shading, 1.0);
}
When I change the return value of get_color to basecolor and disable the normal mapping, the contents in the two buffers are the same. Thus, I am sure the inconsistency is caused by the p3d_NormalMatrix and tbn matrix in the get_color function.
I am not good at CG. But I guess when shading the scene with the second camera, the p3d_NormalMatrix is calculated with the showbase.camera thus raising the problem. Actually, I verified my guess by synchronizing the pose of both cameras in every frame. It turns out the scene in two buffers is the same if the two cameras are at the same position with the same Hpr.
Could you confirm my guess? If it is true, how can I fix this? Thank you so much.
