Merge pull request #56 from pierotofy/orient

Preserve scale/orientation of scene input

colmap.cpp

@@ -127,12 +127,11 @@ InputData inputDataFromColmap(const std::string &projectRoot){
     }
 
     imgf.close();
-    
-    auto r = autoOrientAndCenterPoses(unorientedPoses);
+
+    auto r = autoScaleAndCenterPoses(unorientedPoses);
     torch::Tensor poses = std::get<0>(r);
-    ret.transformMatrix = std::get<1>(r);
-    ret.scaleFactor = 1.0f / torch::max(torch::abs(poses.index({Slice(), Slice(None, 3), 3}))).item<float>();
-    poses.index({Slice(), Slice(None, 3), 3}) *= ret.scaleFactor;
+    ret.translation = std::get<1>(r);
+    ret.scale = std::get<2>(r);
 
     for (size_t i = 0; i < ret.cameras.size(); i++){
         ret.cameras[i].camToWorld = poses[i];
@@ -141,9 +140,7 @@ InputData inputDataFromColmap(const std::string &projectRoot){
     PointSet *pSet = readPointSet(pointsPath.string());
     torch::Tensor points = pSet->pointsTensor().clone();
 
-    ret.points.xyz = torch::matmul(torch::cat({points, torch::ones_like(points.index({"...", Slice(None, 1)}))}, -1), 
-                    ret.transformMatrix.transpose(0, 1));
-    ret.points.xyz *= ret.scaleFactor;
+    ret.points.xyz = (points - ret.translation) * ret.scale;
     ret.points.rgb = pSet->colorsTensor().clone();
 
     RELEASE_POINTSET(pSet);

input_data.hpp

@@ -52,8 +52,8 @@ struct Points{
 };
 struct InputData{
     std::vector<Camera> cameras;
-    float scaleFactor;
-    torch::Tensor transformMatrix;
+    float scale;
+    torch::Tensor translation;
     Points points;
 
     std::tuple<std::vector<Camera>, Camera *> getCameras(bool validate, const std::string &valImage = "random");

model.cpp

@@ -498,10 +498,10 @@ void Model::savePlySplat(const std::string &filename){
 
     float zeros[] = { 0.0f, 0.0f, 0.0f };
 
-    torch::Tensor meansCpu = means.cpu();
+    torch::Tensor meansCpu = (means.cpu() / scale) + translation;
     torch::Tensor featuresDcCpu = featuresDc.cpu();
     torch::Tensor opacitiesCpu = opacities.cpu();
-    torch::Tensor scalesCpu = scales.cpu();
+    torch::Tensor scalesCpu = torch::log((torch::exp(scales.cpu()) / scale));
     torch::Tensor quatsCpu = quats.cpu();
 
     for (size_t i = 0; i < numPoints; i++) {
@@ -518,6 +518,35 @@ void Model::savePlySplat(const std::string &filename){
     std::cout << "Wrote " << filename << std::endl;
 }
 
+void Model::saveDebugPly(const std::string &filename){
+    // A standard PLY
+    std::ofstream o(filename, std::ios::binary);
+    int numPoints = means.size(0);
+
+    o << "ply" << std::endl;
+    o << "format binary_little_endian 1.0" << std::endl;
+    o << "comment Generated by opensplat" << std::endl;
+    o << "element vertex " << numPoints << std::endl;
+    o << "property float x" << std::endl;
+    o << "property float y" << std::endl;
+    o << "property float z" << std::endl;
+    o << "property uchar red" << std::endl;
+    o << "property uchar green" << std::endl;
+    o << "property uchar blue" << std::endl;
+    o << "end_header" << std::endl;
+
+    torch::Tensor meansCpu = (means.cpu() / scale) + translation;
+    torch::Tensor rgbsCpu = (sh2rgb(featuresDc.cpu()) * 255.0f).toType(torch::kUInt8);
+
+    for (size_t i = 0; i < numPoints; i++) {
+        o.write(reinterpret_cast<const char *>(meansCpu[i].data_ptr()), sizeof(float) * 3);
+        o.write(reinterpret_cast<const char *>(rgbsCpu[i].data_ptr()), sizeof(uint8_t) * 3);
+    }
+
+    o.close();
+    std::cout << "Wrote " << filename << std::endl;
+}
+
 torch::Tensor Model::mainLoss(torch::Tensor &rgb, torch::Tensor &gt, float ssimWeight){
     torch::Tensor ssimLoss = 1.0f - ssim.eval(rgb, gt);
     torch::Tensor l1Loss = l1(rgb, gt);

model.hpp

@@ -20,7 +20,7 @@ torch::Tensor psnr(const torch::Tensor& rendered, const torch::Tensor& gt);
 torch::Tensor l1(const torch::Tensor& rendered, const torch::Tensor& gt);
 
 struct Model{
-  Model(const Points &points, int numCameras,
+  Model(const InputData &inputData, int numCameras,
         int numDownscales, int resolutionSchedule, int shDegree, int shDegreeInterval, 
         int refineEvery, int warmupLength, int resetAlphaEvery, int stopSplitAt, float densifyGradThresh, float densifySizeThresh, int stopScreenSizeAt, float splitScreenSize,
         int maxSteps,
@@ -30,17 +30,21 @@ struct Model{
     refineEvery(refineEvery), warmupLength(warmupLength), resetAlphaEvery(resetAlphaEvery), stopSplitAt(stopSplitAt), densifyGradThresh(densifyGradThresh), densifySizeThresh(densifySizeThresh), stopScreenSizeAt(stopScreenSizeAt), splitScreenSize(splitScreenSize),
     maxSteps(maxSteps),
     device(device), ssim(11, 3){
-    long long numPoints = points.xyz.size(0); 
+
+    long long numPoints = inputData.points.xyz.size(0);
+    scale = inputData.scale;
+    translation = inputData.translation;
+
     torch::manual_seed(42);
 
-    means = points.xyz.to(device).requires_grad_();
-    scales = PointsTensor(points.xyz).scales().repeat({1, 3}).log().to(device).requires_grad_();
+    means = inputData.points.xyz.to(device).requires_grad_();
+    scales = PointsTensor(inputData.points.xyz).scales().repeat({1, 3}).log().to(device).requires_grad_();
     quats = randomQuatTensor(numPoints).to(device).requires_grad_();
 
     int dimSh = numShBases(shDegree);
     torch::Tensor shs = torch::zeros({numPoints, dimSh, 3}, torch::TensorOptions().dtype(torch::kFloat32).device(device));
 
-    shs.index({Slice(), 0, Slice(None, 3)}) = rgb2sh(points.rgb.toType(torch::kFloat64) / 255.0).toType(torch::kFloat32);
+    shs.index({Slice(), 0, Slice(None, 3)}) = rgb2sh(inputData.points.rgb.toType(torch::kFloat64) / 255.0).toType(torch::kFloat32);
     shs.index({Slice(), Slice(1, None), Slice(3, None)}) = 0.0f;
 
     featuresDc = shs.index({Slice(), 0, Slice()}).to(device).requires_grad_();
@@ -78,6 +82,7 @@ struct Model{
   int getDownscaleFactor(int step);
   void afterTrain(int step);
   void savePlySplat(const std::string &filename);
+  void saveDebugPly(const std::string &filename);
   torch::Tensor mainLoss(torch::Tensor &rgb, torch::Tensor &gt, float ssimWeight);
 
   void addToOptimizer(torch::optim::Adam *optimizer, const torch::Tensor &newParam, const torch::Tensor &idcs, int nSamples);
@@ -126,6 +131,9 @@ struct Model{
   int stopScreenSizeAt;
   float splitScreenSize;
   int maxSteps;
+
+  float scale;
+  torch::Tensor translation;
 };
 
 

nerfstudio.cpp

@@ -136,12 +136,10 @@ InputData inputDataFromNerfStudio(const std::string &projectRoot){
 
     torch::Tensor unorientedPoses = posesFromTransforms(t);
 
-    auto r = autoOrientAndCenterPoses(unorientedPoses);
+    auto r = autoScaleAndCenterPoses(unorientedPoses);
     torch::Tensor poses = std::get<0>(r);
-    ret.transformMatrix = std::get<1>(r);
-
-    ret.scaleFactor = 1.0f / torch::max(torch::abs(poses.index({Slice(), Slice(None, 3), 3}))).item<float>();
-    poses.index({Slice(), Slice(None, 3), 3}) *= ret.scaleFactor;
+    ret.translation = std::get<1>(r);
+    ret.scale = std::get<2>(r);
 
     // aabbScale = [[-1.0, -1.0, -1.0], [1.0, 1.0, 1.0]]
 
@@ -158,10 +156,8 @@ InputData inputDataFromNerfStudio(const std::string &projectRoot){
     }
 
     torch::Tensor points = pSet->pointsTensor().clone();
-
-    ret.points.xyz = torch::matmul(torch::cat({points, torch::ones_like(points.index({"...", Slice(None, 1)}))}, -1), 
-                    ret.transformMatrix.transpose(0, 1));
-    ret.points.xyz *= ret.scaleFactor;
+
+    ret.points.xyz = (points - ret.translation) * ret.scale;
     ret.points.rgb = pSet->colorsTensor().clone();
 
     RELEASE_POINTSET(pSet);

opensplat.cpp

@@ -101,12 +101,12 @@ int main(int argc, char *argv[]){
         std::vector<Camera> cams = std::get<0>(t);
         Camera *valCam = std::get<1>(t);
 
-        Model model(inputData.points, 
-                        cams.size(),
-                        numDownscales, resolutionSchedule, shDegree, shDegreeInterval, 
-                        refineEvery, warmupLength, resetAlphaEvery, stopSplitAt, densifyGradThresh, densifySizeThresh, stopScreenSizeAt, splitScreenSize,
-                        numIters,
-                        device);
+        Model model(inputData,
+                    cams.size(),
+                    numDownscales, resolutionSchedule, shDegree, shDegreeInterval, 
+                    refineEvery, warmupLength, resetAlphaEvery, stopSplitAt, densifyGradThresh, densifySizeThresh, stopScreenSizeAt, splitScreenSize,
+                    numIters,
+                    device);
 
         std::vector< size_t > camIndices( cams.size() );
         std::iota( camIndices.begin(), camIndices.end(), 0 );
@@ -145,6 +145,7 @@ int main(int argc, char *argv[]){
         }
 
         model.savePlySplat(outputScene);
+        // model.saveDebugPly("debug.ply");
 
         // Validate
         if (valCam != nullptr){

spherical_harmonics.cpp

@@ -15,12 +15,18 @@ int degFromSh(int numBases){
     }
 }
 
+const double C0 = 0.28209479177387814;
+
 torch::Tensor rgb2sh(const torch::Tensor &rgb){
     // Converts from RGB values [0,1] to the 0th spherical harmonic coefficient
-    const double C0 = 0.28209479177387814;
     return (rgb - 0.5) / C0;
 }
 
+torch::Tensor sh2rgb(const torch::Tensor &sh){
+    // Converts from 0th spherical harmonic coefficients to RGB values [0,1]
+    return (sh * C0) + 0.5;
+}
+
 #if defined(USE_HIP) || defined(USE_CUDA)
 
 torch::Tensor SphericalHarmonics::forward(AutogradContext *ctx, 

spherical_harmonics.hpp

@@ -8,6 +8,7 @@ using namespace torch::autograd;
 
 int degFromSh(int numBases);
 torch::Tensor rgb2sh(const torch::Tensor &rgb);
+torch::Tensor sh2rgb(const torch::Tensor &sh);
 
 #if defined(USE_HIP) || defined(USE_CUDA)
 

tensor_math.cpp

@@ -28,17 +28,20 @@ torch::Tensor quatToRotMat(const torch::Tensor &quat){
 
 }
 
-std::tuple<torch::Tensor, torch::Tensor> autoOrientAndCenterPoses(const torch::Tensor &poses){
-    // Center at mean and orient up
+std::tuple<torch::Tensor, torch::Tensor, float> autoScaleAndCenterPoses(const torch::Tensor &poses){
+    // Center at mean
     torch::Tensor origins = poses.index({"...", Slice(None, 3), 3});
-    torch::Tensor translation = torch::mean(origins, 0);
-    torch::Tensor up = torch::mean(poses.index({Slice(), Slice(None, 3), 1}), 0);
-    up = up / up.norm();
+    torch::Tensor center = torch::mean(origins, 0);
+    origins -= center;
+
+    // Scale
+    float f = 1.0f / torch::max(torch::abs(origins)).item<float>();
+    origins *= f;
 
-    torch::Tensor rotation = rotationMatrix(up, torch::tensor({0, 0, 1}, torch::kFloat32));
-    torch::Tensor transform = torch::cat({rotation, torch::matmul(rotation, -translation.index({"...", None}))}, -1);
-    torch::Tensor orientedPoses = torch::matmul(transform, poses);
-    return std::make_tuple(orientedPoses, transform);
+    torch::Tensor transformedPoses = poses.clone();
+    transformedPoses.index_put_({"...", Slice(None, 3), 3}, origins);
+
+    return std::make_tuple(transformedPoses, center, f);
 }
 
 

tensor_math.hpp

@@ -5,7 +5,7 @@
 #include <tuple>
 
 torch::Tensor quatToRotMat(const torch::Tensor &quat);
-std::tuple<torch::Tensor, torch::Tensor> autoOrientAndCenterPoses(const torch::Tensor &poses);
+std::tuple<torch::Tensor, torch::Tensor, float> autoScaleAndCenterPoses(const torch::Tensor &poses);
 torch::Tensor rotationMatrix(const torch::Tensor &a, const torch::Tensor &b);