diff --git a/pages/CSE332S/CSE332S_L17.md b/pages/CSE332S/CSE332S_L17.md
new file mode 100644
index 0000000..7e96681
--- /dev/null
+++ b/pages/CSE332S/CSE332S_L17.md
@@ -0,0 +1,147 @@
+# CSE332S Lecture 17
+
+## Object Oriented Programming Building Blocks
+
+OOP Building Blocks for Extensible, Flexible, and Reusable Code
+
+Today: Techniques Commonly Used in Design Patterns
+
+- **Program to an interface** (last time)
+- **Object composition and request forwarding** (today)
+  - Composition vs. inheritance
+- **Run-time relationships between objects** (today)
+  - Aggregate vs. acquaintance
+- **Delegation** (later...)
+
+Next Time: Design Patterns
+
+Describe the core of a repeatable solution to common design problems.
+
+### Code Reuse: Two Ways to Reuse a Class
+
+#### Inheritance
+
+Code reuse by inheriting the implementation of a base class.
+
+- **Pros:**
+  - Inheritance relationships defined at compile-time - simple to understand.
+- **Cons:**
+  - Subclass often inherits some implementation from superclass - derived class now depends on its base class implementation, leading to less flexible code.
+
+#### Composition
+
+Assemble multiple objects together to create new complex functionality, forward requests to the responsible assembled object.
+
+- **Pros:**
+  - Allows flexibility at run-time, composite objects often constructed dynamically by obtaining references/pointers to other objects (dependency injection).
+  - Objects known only through their interface - increased flexibility, reduced impact of change.
+- **Cons:**
+  - Code can be more difficult to understand, how objects interact may change dynamically.
+
+### Example: Our First Design Pattern (Adapter Pattern)
+
+**Problem:** We are given a class that we cannot modify for some reason - it provides functionality we need, but defines an interface that does not match our program (client code).
+
+**Solution:** Create an adapter class, adapter declares the interface needed by our program, defines it by forwarding requests to the unmodifiable object.
+
+Two ways to do this:
+
+```cpp
+class unmodifiable {
+public:
+    int func(); // does something useful, but doesn’t match the interface required by the client code
+};
+```
+
+1. **Inheritance**
+
+   ```cpp
+   // Using inheritance:
+   class adapter : protected unmodifiable {
+    // open the access to the protected member func() for derived class
+   public:
+       int myFunc() {
+           return func(); // forward request to encapsulated object
+       }
+   };
+   ```
+
+2. **Composition**
+
+   ```cpp
+   class adapterComp {
+       unmodifiable var;
+   public:
+       int myFunc() {
+           return var.func();
+       }
+   };
+   ```
+
+### Thinking About and Describing Run-time Relationships
+
+Typically, composition is favored over inheritance! Object composition with programming to an interface allows relationships/interactions between objects to vary at run-time.
+
+- **Aggregate:** Object is part of another. Its lifetime is the same as the object it is contained in. (similar to base class and derived class relationship)
+- **Acquaintance:** Objects know of each other, but are not responsible for each other. Lifetimes may be different.
+
+```cpp
+// declare Printable Interface
+// declare printable interface
+class printable {
+    public:
+    virtual void print(ostream &o) = 0;
+};
+// derived classes defines printable
+// interface
+class smiley : public printable {
+    public:
+    virtual void print(ostream &o) {
+        o << ":)";
+    };
+};
+// second derived class defines
+// printable interface
+class frown : public printable {
+    public:
+        virtual void print(ostream &o) {o << ":(";
+    };
+};
+```
+
+1. **Aggregate**
+
+```cpp
+// implementation 1:
+// Aggregate relationship
+class emojis {
+printable * happy;
+printable * sad;
+public:
+emojis() {
+happy = new smiley();
+sad = new frown();
+};
+~emojis() {
+delete happy;
+delete sad;
+};
+};
+```
+
+2. **Acquaintance**
+
+```cpp
+// implementation 2:
+// Acquaintances only
+class emojis {
+printable * happy;
+printable * sad;
+public:
+emojis();
+~emojis();
+// dependency injection
+void setHappy(printable *);
+void setSad(printable *);
+};
+```
diff --git a/pages/CSE559A/CSE559A_L19.md b/pages/CSE559A/CSE559A_L19.md
new file mode 100644
index 0000000..ad84824
--- /dev/null
+++ b/pages/CSE559A/CSE559A_L19.md
@@ -0,0 +1,71 @@
+# CSE559A Lecture 19
+
+## Feature Detection
+
+### Behavior of corner features with respect to Image Transformations
+
+To be useful for image matching, “the same” corner features need to show up despite geometric and photometric transformations
+
+We need to analyze how the corner response function and the corner locations change in response to various transformations
+
+#### Affine intensity change
+
+Solution:
+
+- Only derivative of intensity are used (invariant to intensity change)
+- Intensity scaling
+
+#### Image translation
+
+Solution:
+
+- Derivatives and window function are shift invariant
+
+#### Image rotation
+
+Second moment ellipse rotates but its shape (i.e. eigenvalues) remains the same
+
+#### Scaling
+
+Classify edges instead of corners
+
+## Automatic Scale selection for interest point detection
+
+### Scale space
+
+We want to extract keypoints with characteristic scales that are equivariant (or covariant) with respect to scaling of the image
+
+Approach: compute a scale-invariant response function over neighborhoods centered at each location $(x,y)$ and a range of scales $\sigma$, find scale-space locations $(x,y,\sigma)$ where this function reaches a local maximum
+
+A particularly convenient response function is given by the scale-normalized Laplacian of Gaussian (LoG) filter:
+
+$$
+\nabla^2_{norm}=\sigma^2\nabla^2\left(\frac{\partial^2}{\partial x^2}g+\frac{\partial^2}{\partial y^2}g\right)
+$$
+
+![Visualization of LoG](./images/Laplacian_of_Gaussian.png)
+
+#### Edge detection with LoG
+
+![Edge detection with LoG](./images/Edge_detection_with_LoG.png)
+
+#### Blob detection with LoG
+
+![Blob detection with LoG](./images/Blob_detection_with_LoG.png)
+
+### Difference of Gaussians (DoG)
+
+DoG has a little more flexibility, since you can select the scales of the Gaussians.
+
+### Scale-invariant feature transform (SIFT)
+
+The main goal of SIFT is to enable image matching in the presence of significant transformations
+
+- To recognize the same keypoint in multiple images, we need to match appearance descriptors or "signatures" in their neighborhoods
+- Descriptors that are locally invariant w.r.t. scale and rotation can handle a wide range of global transformations
+
+### Maximum stable extremal regions (MSER)
+
+Based on Watershed segmentation algorithm
+
+Select regions that are stable over a large parameter range 
diff --git a/pages/CSE559A/_meta.js b/pages/CSE559A/_meta.js
index 3a518e5..1d8dc77 100644
--- a/pages/CSE559A/_meta.js
+++ b/pages/CSE559A/_meta.js
@@ -21,4 +21,5 @@ export default {
     CSE559A_L16: "Computer Vision (Lecture 16)",
     CSE559A_L17: "Computer Vision (Lecture 17)",
     CSE559A_L18: "Computer Vision (Lecture 18)",
+    CSE559A_L19: "Computer Vision (Lecture 19)",
 }
diff --git a/pages/Math4121/Math4121_L29.md b/pages/Math4121/Math4121_L29.md
index c21aece..08ff5b9 100644
--- a/pages/Math4121/Math4121_L29.md
+++ b/pages/Math4121/Math4121_L29.md
@@ -1 +1,118 @@
-# Lecture 29
\ No newline at end of file
+# Math4121 Lecture 29
+
+## Continue on Measure Theory
+
+### Lebesgue Measure
+
+Caratheodory's criterion:
+
+$S$ is Lebesgue measurable if for all $A\subset S$,
+
+$$
+m_e(X) = m_e(X\cap S) + m_e(X\cap S^c)
+$$
+
+Let $\mathfrak{M}$ be the collection of all Lebesgue measurable sets.
+
+1. $\phi\in\mathfrak{M}$
+2. $\mathfrak{M}$ is closed under countable unions (proved last lecture)
+3. $\mathfrak{M}$ is closed under complementation ($\mathfrak{M}$ is a $\sigma$-algebra) (goal today)
+
+> Desired properties of a measure:
+>
+> 1. $m(I)=\ell(I)$ for all intervals $I$
+> 2. If $\{S_n\}_{n=1}^{\infty}$ is a set of pairwise disjoint Lebesgue measurable sets, then
+>
+> $$ m\left(\bigcup_{n=1}^{\infty}S_n\right) = \sum_{n=1}^{\infty}m(S_n)$$
+> 3. If $R\subset S$, then $m(S\setminus R) = m(S) - m(R)$
+
+Recall the Borel $\sigma$-algebra $\mathcal{B}$ was the smallest $\sigma$-algebra containing closed intervals. Therefore $\mathcal{B}\subset\mathfrak{M}$.
+
+Towards proving $\mathfrak{M}$ is closed under countable unions:
+
+#### Theorem 5.9 (Finite union/intersection of Lebesgue measurable sets is Lebesgue measurable)
+
+Any finite union/intersection of Lebesgue measurable sets is Lebesgue measurable.
+
+Proof:
+
+Suppose $S_1, S_2$ is a measurable, and we need to show that $S_1\cup S_2$ is measurable. Given $X$, need to show that
+
+![Finite union cut](https://notenextra.trance-0.com/math4121/Finite_union_cut.png)
+
+$$
+m_e(X) = m_e(X_1\cup X_2\cup X_3)+ m_e(X_4)
+$$
+
+Since $S_1$ measurable, $m_e(X_1\cup X_2\cup X_3)=m_e(X_3)+m_e(X_1\cup X_2)$.
+
+Since $S_2$ measurable, $m_e(X_3\cup X_4)=m_e(X_3)+m_e(X_4)$.
+
+Therefore,
+
+$$
+\begin{aligned}
+m_e(X) &= m_e(X_1\cup X_2\cup X_3) + m_e(X_4) \\
+&= m_e(X_1\cup X_2) + m_e(X_3)+m_e(X_4) \\
+&= m_e(X_1\cup X_2) + m_e(X_3\cup X_4) \\
+&= m_e(X)
+\end{aligned}
+$$
+
+by measurability of $S_1$ again.
+
+QED
+
+#### Theorem 5.10 (Countable union/intersection of Lebesgue measurable sets is Lebesgue measurable)
+
+Any countable union/intersection of Lebesgue measurable sets is Lebesgue measurable.
+
+Proof:
+
+Let $\{S_j\}_{j=1}^{\infty}\subset\mathfrak{M}$. Definte $T_j=\bigcup_{k=1}^{j}S_k$ such that $T_{j-1}\subset T_j$ for all $j$.
+
+And $U_1=T_1$, $U_j=T_j\setminus T_{j-1}$ for $j\geq 2$.
+
+Then $\bigcup_{j=1}^{\infty}S_j=\bigcup_{j=1}^{\infty}T_j=\bigcup_{j=1}^{\infty}U_j$. Notice that $\{U_j\}_{j=1}^{\infty}$ are pairwise disjoint, and $\{T_j\}_{j=1}^{\infty}$ are monotone.
+
+Let $X$ have finite outer measure. Since $U_n$ is measurable,
+
+$$
+\begin{aligned}
+m_e(X\cap T_n) &= m_e(X\cap T_n\cap U_n)+ m_e(X\cap T_n\cap U_n^c) \\
+&= m_e(X\cap U_n)+ m_e(X\cap T_{n-1}) \\
+&= \sum_{j=1}^{n}m_e(X\cap U_j)
+\end{aligned}
+$$
+
+Since $T_n$ is measurable and $T_n\subset S$, $S^c\subset T_n^c$. $m_e(X\cap T_n^c)\geq m_e(X\cap S^c)$.
+
+Therefore,
+
+$$
+m_e(X)=m_e(X\cap T_n)+m_e(X\cap T_n^c)\\
+\geq \sum_{j=1}^{n}m_e(X\cap U_j)+m_e(X\cap S^c)
+$$
+
+Take the limit as $n\to\infty$,
+
+$$
+\begin{aligned}
+m_e(X) &\geq \sum_{j=1}^{\infty}m_e(X\cap U_j)+m_e(X\cap S^c) \\
+&= m_e(\bigcup_{j=1}^{\infty}(X\cap U_j))+m_e(X\cap S^c) \\
+&= m_e(X\cap S)+m_e(X\cap S^c) \\
+&\geq m_e(X)
+\end{aligned}
+$$
+
+Therefore, $m_e(X\cap S)=m_e(X)$.
+
+Therefore, $S$ is measurable.
+
+QED
+
+#### Corollary from the proof
+
+Every open or closed set is Lebesgue measurable.
+
+(Every open set is a countable union of disjoint open intervals)
diff --git a/public/Math4121/Finite_union_cut.png b/public/Math4121/Finite_union_cut.png
new file mode 100644
index 0000000..0f4e997
Binary files /dev/null and b/public/Math4121/Finite_union_cut.png differ