diff --git a/research/DSG/index.html b/research/DSG/index.html
index cf5833f6..8925073d 100644
--- a/research/DSG/index.html
+++ b/research/DSG/index.html
@@ -781,6 +781,15 @@ Geodesic Finite Differences (GFD)
$$J_{Exp}^v \approx M_{p'}^{\dagger} \cdot \left[ \frac{Exp(p, v+\epsilon_v \hat{e}_{||}) - p'}{\epsilon_v} \quad \frac{Exp(p, v+\epsilon_v \hat{e}_{\perp}) - p'}{\epsilon_v} \right]$$
+ GPU Parallelisation
+
+ We implement both methods in CUDA, parallelising the tracing of multiple geodesics across the mesh. This allows us to efficiently compute the exponential map and its Jacobians for large batches of points and vectors, making it practical for integration into deep learning pipelines.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/research/DSG/static/images/benchmark_facecount.svg b/research/DSG/static/images/benchmark_facecount.svg
new file mode 100644
index 00000000..1944fffc
--- /dev/null
+++ b/research/DSG/static/images/benchmark_facecount.svg
@@ -0,0 +1,304 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ 
+