Benjamin Martinez

I am a 4th year undergrad student at Stanford University studying computer science with a concentration on visual computing. This concentration covers my interests in machine learning, graphics, computer vision, and virtual reality. I currently work at the Stanford Vision and Learning Lab working on the BEHAVIOR-1K project. I plan to pursue a master's in AI.

Broader interests include philosophy, mysticism, language, film, animation, video games, mindfulness, and surfing.

Email  /  Resume  /  Linkedin

Projects

Select things I've worked on for classes, internships, or in my free time.

BEHAVIOR is a human-centered simulation benchmark where embodied AI agents need to plan and execute navigation and manipulation strategies based on sensor information to fulfill up to 1,000 household activities across 50 scenes. BEHAVIOR tests the ability of agents to perceive the environment, plan, and execute complex long-horizon activities that involve multiple objects, rooms, and state changes, all with the reproducibility, safety, and observability offered by a realistic physics simulation.

Machine Learning

Our team in CS231N designed a novel method for optimized data selection using convex optimization to diagnose chest X-ray images with convolutional neural networks

CS 229 project to analyze and experiment with state-of-the-art meta-learning model weight initialization procedures. Co-developed a novel method to optimize training by shaping loss curves.

PSYC 221 project to identify brain areas correlated with ASD social disorders by using KNNs on dynamic functional connectivity patterns in time series fMRIs

Computer Graphics

Implemented editor with basic mesh edit operations such as beveling, collapsing edges, triangulation, subdivision, and simplification.

Implemented efficient ray-scene geometry queries, bounding volume hierarchies, materials, and path tracing for global illumination.

Bare metal programmed a Raspberry Pi to emulate the original Gameboy hardware. Used C to build a graphics library, allocate memory, and process CPU instructions from the ground up

GLSL shader programming for material pattern logic, material BRDFs, and lighting computations to compute surface reflectance. Implemented several lighting models including Phong, Cook-Torrance, and Ward.

Simulate 2D deformable “blobs” and process collisions between them in a rigid environment.

Built an interactive inverse kinematics game in PyBullet that mimics the Super Hot game.

3D modeled kirby with ray marching and signed distance fields in GLSL

Generative Models

Here I study, reimplement, or experiment with papers or ideas in 3D geometry. Note: These are not original contributions these are for my education.

I self studied NeRF and reimplemented the model to predict a depth map rather than color.

Implemented a vanilla PointNet, training an autoencoder and variational autoencoder for reconstructing 3D shapes in point cloud representation

Implemented model trained to learn signed distance field (SDF) of input bunny point cloud. Then implemented the marching cubes algorithm to extract the mesh from the SDF.

Adapted from Jon Barron's website.