add stb_image_resize2.h

implement some Display P3 color space conversion stuff.

MANIFEST.in

@@ -175,6 +175,7 @@ include src/external/simple_match/include/simple_match/simple_match.hpp
 include src/external/simple_match/include/simple_match/utility.hpp
 include src/external/stb_image.h
 include src/external/stb_image_resize.h
+include src/external/stb_image_resize2.h
 include src/external/stb_image_write.h
 include src/external/stb_truetype.h
 include src/external/string_id/LICENSE

examples/usd_to_gltf/to-gltf-main.cc

@@ -49,6 +49,46 @@ using PrimvarReader_float2Map =
     std::map<std::string, std::pair<const tinyusdz::Shader *,
                                     const tinyusdz::UsdPrimvarReader_float2 *>>;
 
+tinygltf::Material to_gltf_material(const tinyusdz::tydra::RenderMaterial &mat) {
+  tinygltf::Material out;
+
+  out.pbrMetallicRoughness.roughnessFactor = mat.surfaceShader.roughness.value;
+
+  return out;
+}
+
+bool to_gltf(const tinyusdz::tydra::RenderScene &rscene, const std::string &gltf_filename)
+{
+  tinygltf::Model model;
+  tinygltf::Scene scene;
+  std::vector<tinygltf::Mesh> meshes;
+  tinygltf::Primitive primitive;
+
+  tinygltf::Asset asset;
+  asset.version = "2.0";
+  asset.generator = "usd_to_gltf example in TinyUSDZ";
+
+  model.scenes.push_back(scene);
+
+  // model.bufferViews
+  // model.buffers
+  // model.nodes
+  //model.meshes = meshes;
+  model.asset = asset;
+
+  // model.materials
+
+  tinygltf::TinyGLTF ctx;
+  bool ret = ctx.WriteGltfSceneToFile(&model,
+    gltf_filename,
+    true, // embedImages
+  true, // embedBuffers
+  true, // pretty print
+  false); // write binary glTF
+
+  return ret;
+}
+
 int main(int argc, char **argv) {
   if (argc < 2) {
     std::cout << "Need USD file.\n"
@@ -158,5 +198,10 @@ int main(int argc, char **argv) {
 
   std::cout << DumpRenderScene(render_scene) << "\n";
 
+  if (!to_gltf(render_scene, "output.gltf")) {
+    std::cerr << "Failed to save scene as glTF\n";
+    return EXIT_FAILURE;
+  } 
+
   return EXIT_SUCCESS;
 }

src/image-util.cc

@@ -15,13 +15,15 @@
 #define STB_IMAGE_RESIZE_IMPLEMENTATION
 #endif
 
-#include "external/stb_image_resize.h"
+//#include "external/stb_image_resize.h"
+#include "external/stb_image_resize2.h"
 
 #ifdef __clang__
 #pragma clang diagnostic pop
 #endif
 
 #include "image-util.hh"
+#include "value-types.hh"
 
 #if defined(TINYUSDZ_WITH_COLORIO)
 #include "external/tiny-color-io.h"
@@ -428,7 +430,7 @@ bool srgb_8bit_to_linear_f32(const std::vector<uint8_t> &in_img, size_t width,
 bool srgb_f32_to_linear_f32(const std::vector<float> &in_img, size_t width,
                          size_t height,
                          size_t channels, size_t channel_stride,
-                         std::vector<float> *out_img, float scale_factor, float bias) {
+                         std::vector<float> *out_img, const float scale_factor, const float bias, const float alpha_scale_factor, const float alpha_bias) {
 
   if ((width == 0) ||
     (height == 0) ||
@@ -465,8 +467,7 @@ bool srgb_f32_to_linear_f32(const std::vector<float> &in_img, size_t width,
       // Apply linear conversion.
       for (size_t c = channels; c < channel_stride; c++) {
         size_t idx = channel_stride * width * y + channel_stride * x + c;
-        // TODO: Do we need to apply scale_factor and bias to linear-value(e.g. alpha) channel also?
+        float f = in_img[idx] * alpha_scale_factor + alpha_bias;
-        float f = in_img[idx] * scale_factor + bias;
         (*out_img)[idx] = f;
       }
     }
@@ -579,4 +580,215 @@ bool f32_to_u8_image(const std::vector<float> &in_img, size_t width,
   return true;
 }
 
+bool linear_displayp3_to_linear_sRGB(const std::vector<float> &in_img, size_t width,
+                         size_t height, size_t channels,
+                         std::vector<float> *out_img) {
+
+  // http://endavid.com/index.php?entry=79
+  // https://tech.metail.com/introduction-colour-spaces-dci-p3/
+  
+
+  if (!out_img) {
+    return false;
+  }
+  
+  if (channels > 4) {
+    return false;
+  }
+
+  if ((channels != 3) && (channels != 4)) {
+    return false;
+  }
+
+  if (in_img.size() != (width * height * channels)) {
+    return false;
+  }
+
+  out_img->resize(in_img.size());
+
+  if (channels == 3) {
+    for (size_t y = 0; y < height; y++) {
+      for (size_t x = 0; x < width; x++) {
+        float r, g, b;
+        r = in_img[3 * (y * width + x) + 0]; 
+        g = in_img[3 * (y * width + x) + 1]; 
+        b = in_img[3 * (y * width + x) + 2]; 
+
+        float out_rgb[3];
+        out_rgb[0] = 1.2249f * r - 0.2247f * g;
+        out_rgb[1] = -0.0420f * r + 1.0419f * g;
+        out_rgb[2] = -0.0197f * r - 0.0786f * g + 1.0979f * b;
+
+        // clamp
+        out_rgb[0] = (out_rgb[0] < 0.0f) ? 0.0f : out_rgb[0];
+        out_rgb[1] = (out_rgb[1] < 0.0f) ? 0.0f : out_rgb[1];
+        out_rgb[2] = (out_rgb[2] < 0.0f) ? 0.0f : out_rgb[2];
+
+        (*out_img)[3 * (y * width + x) + 0] = out_rgb[0];
+        (*out_img)[3 * (y * width + x) + 1] = out_rgb[1];
+        (*out_img)[3 * (y * width + x) + 2] = out_rgb[2];
+      }
+    }
+
+  } else { // rgba
+    for (size_t y = 0; y < height; y++) {
+      for (size_t x = 0; x < width; x++) {
+        float r, g, b, a;
+        r = in_img[4 * (y * width + x) + 0]; 
+        g = in_img[4 * (y * width + x) + 1]; 
+        b = in_img[4 * (y * width + x) + 2]; 
+        a = in_img[4 * (y * width + x) + 3]; 
+
+        float out_rgb[3];
+        out_rgb[0] = 1.2249f * r - 0.2247f * g;
+        out_rgb[1] = -0.0420f * r + 1.0419f * g;
+        out_rgb[2] = -0.0197f * r - 0.0786f * g + 1.0979f * b;
+
+        // clamp
+        out_rgb[0] = (out_rgb[0] < 0.0f) ? 0.0f : out_rgb[0];
+        out_rgb[1] = (out_rgb[1] < 0.0f) ? 0.0f : out_rgb[1];
+        out_rgb[2] = (out_rgb[2] < 0.0f) ? 0.0f : out_rgb[2];
+
+        (*out_img)[4 * (y * width + x) + 0] = out_rgb[0];
+        (*out_img)[4 * (y * width + x) + 1] = out_rgb[1];
+        (*out_img)[4 * (y * width + x) + 2] = out_rgb[2];
+        (*out_img)[4 * (y * width + x) + 3] = a;
+      }
+    }
+  }
+
+  return true;
+}
+
+bool linear_sRGB_to_linear_displayp3(const std::vector<float> &in_img, size_t width,
+                         size_t height, size_t channels,
+                         std::vector<float> *out_img) {
+
+  // http://endavid.com/index.php?entry=79
+  // https://tech.metail.com/introduction-colour-spaces-dci-p3/
+  
+
+  if (!out_img) {
+    return false;
+  }
+  
+  if (channels > 4) {
+    return false;
+  }
+
+  if ((channels != 3) && (channels != 4)) {
+    return false;
+  }
+
+  if (in_img.size() != (width * height * channels)) {
+    return false;
+  }
+
+  out_img->resize(in_img.size());
+
+  if (channels == 3) {
+    for (size_t y = 0; y < height; y++) {
+      for (size_t x = 0; x < width; x++) {
+        float r, g, b;
+        r = in_img[3 * (y * width + x) + 0]; 
+        g = in_img[3 * (y * width + x) + 1]; 
+        b = in_img[3 * (y * width + x) + 2]; 
+
+        float out_rgb[3];
+        out_rgb[0] = 0.8225f * r + 0.1774f * g;
+        out_rgb[1] = 0.0332f * r + 0.9669f * g;
+        out_rgb[2] = 0.0171f * r + 0.0724f * g + 0.9108f * b;
+
+        // clamp for just in case.
+        out_rgb[0] = (out_rgb[0] < 0.0f) ? 0.0f : out_rgb[0];
+        out_rgb[1] = (out_rgb[1] < 0.0f) ? 0.0f : out_rgb[1];
+        out_rgb[2] = (out_rgb[2] < 0.0f) ? 0.0f : out_rgb[2];
+
+        (*out_img)[3 * (y * width + x) + 0] = out_rgb[0];
+        (*out_img)[3 * (y * width + x) + 1] = out_rgb[1];
+        (*out_img)[3 * (y * width + x) + 2] = out_rgb[2];
+      }
+    }
+
+  } else { // rgba
+    for (size_t y = 0; y < height; y++) {
+      for (size_t x = 0; x < width; x++) {
+        float r, g, b, a;
+        r = in_img[4 * (y * width + x) + 0]; 
+        g = in_img[4 * (y * width + x) + 1]; 
+        b = in_img[4 * (y * width + x) + 2]; 
+        a = in_img[4 * (y * width + x) + 3]; 
+
+        float out_rgb[3];
+        out_rgb[0] = 0.8225f * r + 0.1774f * g;
+        out_rgb[1] = 0.0332f * r + 0.9669f * g;
+        out_rgb[2] = 0.0171f * r + 0.0724f * g + 0.9108f * b;
+
+        // clamp for just in case.
+        out_rgb[0] = (out_rgb[0] < 0.0f) ? 0.0f : out_rgb[0];
+        out_rgb[1] = (out_rgb[1] < 0.0f) ? 0.0f : out_rgb[1];
+        out_rgb[2] = (out_rgb[2] < 0.0f) ? 0.0f : out_rgb[2];
+
+        (*out_img)[4 * (y * width + x) + 0] = out_rgb[0];
+        (*out_img)[4 * (y * width + x) + 1] = out_rgb[1];
+        (*out_img)[4 * (y * width + x) + 2] = out_rgb[2];
+        (*out_img)[4 * (y * width + x) + 3] = a;
+      }
+    }
+  }
+
+  return true;
+}
+
+bool displayp3_f16_to_linear_f32(const std::vector<value::half> &in_img, size_t width,
+                         size_t height,
+                         size_t channels, size_t channel_stride,
+                         std::vector<float> *out_img, const float scale_factor, const float bias, const float alpha_scale_factor, const float alpha_bias) {
+
+  if ((width == 0) ||
+    (height == 0) ||
+    (channels == 0) ||
+    (out_img == nullptr)) {
+    return false;
+  }
+
+  if (channel_stride == 0) {
+    channel_stride = channels;
+  } else {
+    if (channel_stride < channels) {
+      return false;
+    }
+  }
+
+  size_t dest_size = size_t(width) * size_t(height) * channel_stride;
+  if (dest_size > in_img.size()) {
+    return false;
+  }
+
+  out_img->resize(dest_size);
+
+  // assume input is in [0.0, 1.0]
+  for (size_t y = 0; y < height; y++) {
+    for (size_t x = 0; x < width; x++) {
+      for (size_t c = 0; c < channels; c++) {
+        size_t idx = channel_stride * width * y + channel_stride * x + c;
+        float in_val = value::half_to_float(in_img[idx]);
+        float f = in_val * scale_factor + bias;
+        (*out_img)[idx] = SrgbTransform::srgbToLinear(f); // Display P3 use the same transfer function with sRGB
+      }
+
+      // remainder(usually alpha channel)
+      // Apply linear conversion.
+      for (size_t c = channels; c < channel_stride; c++) {
+        size_t idx = channel_stride * width * y + channel_stride * x + c;
+        float in_val = value::half_to_float(in_img[idx]);
+        float f = in_val * alpha_scale_factor + alpha_bias;
+        (*out_img)[idx] = f;
+      }
+    }
+  }
+
+  return true;
+}
+
 } // namespace tinyusdz

src/image-util.hh

@@ -5,7 +5,7 @@
 // Currently sRGB color space conversion feature is provided.
 //
 // TODO
-// - [ ] Image resize
+// - [ ] Image resize using stb_image_resize2
 // - [ ] OIIO 3D LUT support through tinycolorio
 //
 #pragma once
@@ -16,6 +16,11 @@
 
 namespace tinyusdz {
 
+// forward decl
+namespace value {
+struct half;
+};
+
 ///
 /// [0, 255] => [0.0, 1.0]
 ///
@@ -52,7 +57,7 @@ bool linear_f32_to_srgb_8bit(const std::vector<float> &in_img, size_t width,
                          std::vector<uint8_t> *out_img);
 
 ///
-/// Convert 8bit image in sRGB to fp32 image in linear color space.
+/// Convert 8bit image in sRGB to fp32 image in linear sRGB color space.
 ///
 /// @param[in] in_image Input image in sRGB color space. Image size =
 /// [width_byte_stride, height, channel_stride]
@@ -75,15 +80,16 @@ bool srgb_8bit_to_linear_8bit(const std::vector<uint8_t> &in_img, size_t width,
                          size_t height, size_t channels, size_t channel_stride,
                          std::vector<uint8_t> *out_img);
 
-// Input texel value is transformed as: x' = in_img * scale_factor + bias
+// Input texel value is transformed as: x' = in_img * scale_factor + bias for RGB
+// alpha' = in_img * alpha_scale_factor + alpha_bias for alpha channel.
 bool srgb_f32_to_linear_f32(const std::vector<float> &in_img, size_t width,
                          size_t height, size_t channels, size_t channel_stride,
-                         std::vector<float> *out_img, float scale_factor = 1.0f, float bias = 0.0f);
+                         std::vector<float> *out_img, const float scale_factor = 1.0f, const float bias = 0.0f, const float alpha_scale_factor = 1.0f, const float alpha_bias = 0.0f);
 
 ///
 /// Convert 8bit image in Rec.709 to fp32 image in linear color space.
 ///
-/// @param[in] in_image Input image in Rec.709 color space. Image size =
+/// @param[in] in_img Input image in Rec.709 color space. Image size =
 /// [width_byte_stride, height, channel_stride]
 /// @param[in] width Width pixels
 /// @param[in] width_byte_stride Width byte stride. 0 = Use `width` *
@@ -103,6 +109,90 @@ bool rec709_8bit_to_linear_f32(const std::vector<uint8_t> &in_img, size_t width,
                          size_t channels, size_t channel_stride,
                          std::vector<float> *out_img);
 
+///
+/// Convert fp16 image in Display P3(P3-D65) to fp32 image in linear Display P3 color space.
+/// 
+/// The conversion is identical to sRGB -> linear sRGB, since Display P3 uses same gamma curve(transfer function) with sRGB.
+///
+/// Input value is scaled by x' = x * scale + bias.
+///
+/// @param[in] in_img Input image in Display P3 color space. Image size =
+/// [width_byte_stride, height, channel_stride]
+/// @param[in] width Width pixels
+/// @param[in] height Height pixels
+/// @param[in] chanels Pixel channels to apply conversion. must be less than or
+/// equal to `channel_stride`
+/// @param[in] chanel_stride channel stride. For example, channels=3 and
+/// channel_stride=4 to apply inverse gamma correction to RGB channel but apply
+/// linear conversion to alpha channel for RGBA image.
+/// @param[out] out_img Image in linear Display P3 color space. Image size is same with
+/// `in_image`
+/// @param[in] scale texel scale factor(RGB)
+/// @param[in] bias texel bias(RGB)
+/// @param[in] alpha_scale texel scale factor(alpha)
+/// @param[in] alpha_bias texel bias(alpha)
+///
+/// @return true upon success. false when any parameter is invalid.
+bool displayp3_f16_to_linear_f32(const std::vector<value::half> &in_img, size_t width,
+                         size_t height, size_t channels, size_t channel_stride,
+                         std::vector<float> *out_img, const float scale_factor = 1.0f, const float bias = 0.0f, const float alpha_scale_factor = 1.0f, const float alpha_bias = 0.0f);
+
+///
+/// Convert fp32 image in linear Display P3 color space to 10bit Display P3(10 bit for RGB, 2 bit for alpha, 32bit in total)
+/// Apply gamma curve + quantize values.
+///
+/// Input image must be Mono, LA(Luminance + Alpha), RGB or RGBA.
+///
+/// Monochrome image is converted to RGB
+///
+/// When alpha channel is supplied, alpha value in [0.0, 1.0] is quantized to 2bit(~0.25 = 0, ~0.5 = 1, ~0.75 = 2, 0.75+ = 3)
+///
+/// @param[in] in_img Input image in linear Display P3 color space. 
+/// @param[in] width Width pixels
+/// @param[in] height Height pixels
+/// @param[in] channels 1 = mono, 2 = luminance(mono) + alpha, 3 = RGB, 4 = RGBA
+/// @param[out] out_img Image in linear Display P3 color space. Image size is same with
+/// `in_image`
+bool linear_f32_to_displayp3_u10(const std::vector<float> &in_img, size_t width,
+                         size_t height, size_t channels,
+                         std::vector<uint32_t> *out_img);
+
+///
+/// Convert linear Display P3 color space to linear sRGB color space.
+///
+/// Input image must be RGB or RGBA.
+/// (TODO: Support Mono, Lumi/Alpha)
+///
+/// When alpha channel is supplied, no conversion applied to alpha value.
+///
+/// @param[in] in_img Input image in linear Display P3 color space. 
+/// @param[in] width Width pixels
+/// @param[in] height Height pixels
+/// @param[in] channels 3 = RGB, 4 = RGBA
+/// @param[out] out_img Image in linear sRGB color space. Image size is same with
+/// `in_image`
+bool linear_displayp3_to_linear_sRGB(const std::vector<float> &in_img, size_t width,
+                         size_t height, size_t channels,
+                         std::vector<float> *out_img);
+
+///
+/// Convert linear sRGB color space to linear Display P3 color space.
+///
+/// Input image must be RGB or RGBA.
+/// (TODO: Support Mono, Lumi/Alpha)
+///
+/// When alpha channel is supplied, no conversion applied to alpha value.
+///
+/// @param[in] in_img Input image in linear sRGB color space. 
+/// @param[in] width Width pixels
+/// @param[in] height Height pixels
+/// @param[in] channels 3 = RGB, 4 = RGBA
+/// @param[out] out_img Image in linear Display P3 color space. Image size is same with
+/// `in_image`
+bool linear_sRGB_to_linear_displayp3(const std::vector<float> &in_img, size_t width,
+                         size_t height, size_t channels,
+                         std::vector<float> *out_img);
+
 ///
 /// Resize fp32 image in linear color space.
 ///

src/tydra/render-data.hh

@@ -166,6 +166,7 @@ struct BufferData {
   // TODO: Stride
 };
 
+#if 0 // not used atm
 // glTF-like Attribute
 struct Attribute {
   std::string path;     // Path string in Stage
@@ -173,6 +174,7 @@ struct Attribute {
 
   int64_t buffer_id{-1};  // index to buffer_id
 };
+#endif
 
 // Compound of ComponentType x component
 enum class VertexAttributeFormat {
@@ -1064,7 +1066,7 @@ class RenderScene {
   std::vector<RenderMesh> meshes;
   std::vector<Animation> animations;
   std::vector<BufferData>
-      buffers;  // Various data storage(e.g. primvar texcoords)
+      buffers;  // Various data storage(e.g. texel/image data).
 
   // int64_t default_root_node{-1}; // index to `nodes`. `defaultPrim` in USD
 };
@@ -1462,26 +1464,6 @@ class RenderSceneConverter {
   RenderSceneConverter(const RenderSceneConverter &rhs) = delete;
   RenderSceneConverter(RenderSceneConverter &&rhs) = delete;
 
-  //void set_scene_config(const RenderSceneConverterConfig &config) {
-  //  _scene_config = config;
-  //}
-
-  //void set_mesh_config(const MeshConverterConfig &config) {
-  //  _mesh_config = config;
-  //}
-
-  //void set_material_config(const MaterialConverterConfig &config) {
-  //  _material_config = config;
-  //}
-
-  //void set_asset_resoluition_resolver(AssetResolutionResolver &&rhs) {
-  //  _asset_resolver = std::move(rhs);
-  //}
-
-  //void set_search_paths(const std::vector<std::string> &paths) {
-  //  _asset_resolver.set_search_paths(paths);
-  //}
-
   ///
   /// All-in-one Stage to RenderScene conversion.
   ///
@@ -1651,12 +1633,6 @@ class RenderSceneConverter {
   bool GetSkeletonImpl(const tinyusdz::Prim &prim,
                        const tinyusdz::Skeleton *&out_skeleton);
 
-  //AssetResolutionResolver _asset_resolver;
-  //RenderSceneConverterConfig _scene_config;
-  //MeshConverterConfig _mesh_config;
-  //MaterialConverterConfig _material_config;
-  // const Stage *_stage{nullptr};
-
   void PushInfo(const std::string &msg) { _info += msg; }
   void PushWarn(const std::string &msg) { _warn += msg; }
   void PushError(const std::string &msg) { _err += msg; }