ImportGltf: experimental support for non-uniform scaling

CHANGELOG.md

@@ -8,6 +8,7 @@
 * The Texture/Info spell is now applicable to any string item that ends with ".dds", and it opens a dialog showing information and a preview of the texture.
 * Implemented Transform/Apply for Starfield (internal geometry only).
 * Starfield and Fallout 76 shading fixes.
+* Experimental support for importing glTF models that use anisotropic scaling. Non-uniform scale factors are applied to the rotation matrix, so the model should be fixed manually by applying the transform and recalculating normals and tangents.
 * Optimizations to the texture and material browsers, and to PBR cube map filtering and loading Radiance HDR format files.
 * Fixed the 'Texture/Choose' spell being incorrectly applicable to items that are not texture paths, and the spell can now also be used on Starfield texture set items without expanding the structure.
 * Fixed error on BSFaceFX and BoneModifierExtra blocks.

src/lib/importex/gltf.cpp

@@ -982,10 +982,12 @@ class ImportGltf {
 	const tinygltf::Model &	model;
 	NifModel *	nif;
 	bool	lodEnabled;
+	bool	scaleWarningFlag;
 	std::vector< int >	nodeStack;
 	bool nodeHasMeshes( const tinygltf::Node & node, int d = 0 ) const;
 	static void normalizeFloats( float * p, size_t n, int dataType );
 	template< typename T > bool loadBuffer( std::vector< T > & outBuf, int accessor, int typeRequired );
+	void applyXYZScale( Transform & t, const Vector3 & scale );
 	void loadSkin( const QPersistentModelIndex & index, const tinygltf::Skin & skin );
 	int loadTriangles( const QModelIndex & index, const tinygltf::Primitive & p );
 	void loadSkinnedLODMesh( const QPersistentModelIndex & index, const tinygltf::Primitive & p, int lod );
@@ -996,7 +998,7 @@ class ImportGltf {
 	void loadNode( const QPersistentModelIndex & index, int nodeNum, bool isRoot );
 public:
 	ImportGltf( NifModel * nifModel, const tinygltf::Model & gltfModel, bool enableLOD )
-		: model( gltfModel ), nif( nifModel ), lodEnabled( enableLOD )
+		: model( gltfModel ), nif( nifModel ), lodEnabled( enableLOD ), scaleWarningFlag( false )
 	{
 	}
 	void importModel( const QPersistentModelIndex & iBlock );
@@ -1121,6 +1123,30 @@ template< typename T > bool ImportGltf::loadBuffer( std::vector< T > & outBuf, i
 	return true;
 }
 
+void ImportGltf::applyXYZScale( Transform & t, const Vector3 & scale )
+{
+	float	avgScale = ( scale[0] + scale[1] + scale[2] ) / 3.0f;
+	t.scale = avgScale;
+	FloatVector4	tmp = FloatVector4( scale ) / avgScale;
+	if ( ( ( ( tmp - 1.0f ).absValues() - 0.000001f ).getSignMask() & 0x07 ) == 0x07 )
+		return;
+	if ( !scaleWarningFlag ) {
+		scaleWarningFlag = true;
+		QMessageBox::warning( nullptr, "NifSkope warning",
+								tr( "glTF model uses anisotropic scaling, use Transform/Apply to fix transforms, "
+									"and recalculate normals and tangents" ) );
+	}
+	t.rotation( 0, 0 ) *= tmp[0];
+	t.rotation( 1, 0 ) *= tmp[0];
+	t.rotation( 2, 0 ) *= tmp[0];
+	t.rotation( 0, 1 ) *= tmp[1];
+	t.rotation( 1, 1 ) *= tmp[1];
+	t.rotation( 2, 1 ) *= tmp[1];
+	t.rotation( 0, 2 ) *= tmp[2];
+	t.rotation( 1, 2 ) *= tmp[2];
+	t.rotation( 2, 2 ) *= tmp[2];
+}
+
 void ImportGltf::loadSkin( const QPersistentModelIndex & index, const tinygltf::Skin & skin )
 {
 	QPersistentModelIndex	iSkinBMP = nif->insertNiBlock( "SkinAttach" );
@@ -1204,7 +1230,7 @@ void ImportGltf::loadSkin( const QPersistentModelIndex & index, const tinygltf::
 			Transform	t;
 			Vector3	tmpScale;
 			m.decompose( t.translation, t.rotation, tmpScale );
-			t.scale = ( tmpScale[0] + tmpScale[1] + tmpScale[2] ) / 3.0f;
+			applyXYZScale( t, tmpScale );
 			QModelIndex	iBone = nif->getIndex( iBones, int(i) );
 			if ( iBone.isValid() )
 				t.writeBack( nif, iBone );
@@ -1610,19 +1636,19 @@ void ImportGltf::loadNode( const QPersistentModelIndex & index, int nodeNum, boo
 				const_cast< float * >( m.data() )[i] = float( node.matrix[i] );
 			Vector3	tmpScale;
 			m.decompose( t.translation, t.rotation, tmpScale );
-			t.scale = ( tmpScale[0] + tmpScale[1] + tmpScale[2] ) / 3.0f;
+			applyXYZScale( t, tmpScale );
 		} else {
 			if ( node.rotation.size() >= 4 ) {
 				Quat	r;
 				for ( size_t i = 0; i < 4; i++ )
 					r[(i + 1) & 3] = float( node.rotation[i] );
 				t.rotation.fromQuat( r );
 			}
-			if ( node.scale.size() >= 1 ) {
-				double	s = 0.0;
-				for ( double i : node.scale )
-					s += i;
-				t.scale = float( s / double( int(node.scale.size()) ) );
+			if ( node.scale.size() >= 3 ) {
+				Vector3	tmpScale( float( node.scale[0] ), float( node.scale[1] ), float( node.scale[2] ) );
+				applyXYZScale( t, tmpScale );
+			} else if ( node.scale.size() >= 1 ) {
+				t.scale = float( node.scale[0] );
 			}
 			if ( node.translation.size() >= 3 ) {
 				t.translation[0] = float( node.translation[0] );