Malcolm Hsiu

Mathematics-Computer Science graduate from University of California, San Diego.

Email  /  CV  /  Bio  /  Disclaimer

profile photo

Projects

ML and DL projects I've done.
Domain Randomization for ManiSkill

Adding randomized lighting direction/color and object texture to ManiSkill.

scale_estimate Metric Scale Estimation

E-commerce metadata provides a rich source of image-to-size pairs because Amazon vendors specify the product's dimensions and provide multiple images of their product in different contexts and with varying camera angles, lighting conditions, etc. This makes it possible to learn metric scale estimation by training on e-commerce data like the Amazon Berkeley Objects dataset.
ACM_stylized Stylized ACM

Stylizing the ACM logo using Stable Diffusion and Control Net.
tensorf_render TensoRF

A from-scratch implementation of Tensorial Radiance Fields.
ddpm_sample Denoising Diffusion Probabilistic Model

A from-scratch implementation of a Denoising Diffusion Probabilistic Model, including a U-Net trained on MNIST.
transformer Transformer Decoder

A from-scratch multi-head Transformer decoder implemented with einsum and einops. Data pre-processing based on Andrej Karpathy's GPT tutorial.
stylish_dog Style Transfer

A from-scratch implementation of Gatys style transfer.
vae_samples Variational Autoencoder

A from-scratch implementation of the variational autoencoder, trained on MNIST.
loss_landscape Loss Landscapes of Residual Networks

A from-scratch implementation of ResNet's trained on CIFAR-10/100. Dataloading accelerated by FFCV. Loss landscapes visualized using "filter normalization".

Made using Jon Barron's source code. Long live The Barron!