Linbo Tang

Gödel, Escher, and Bach.

Emo-dynamics: Utilizing Dynamics in Selfie Videos to Capture Emotions, Singh’s CV Group

  • Curating EmotionNet5k, a novel selfie video dataset containing 5000 samples of facial expressions, with over 300 stars on GitHub, expected to be published on ECCV 2024;
  • Implementing a SOTA learnable wavelet model to remove co-founders and achieve a 3.5 % increase in accuracy on the downstream Facial Emotion Recognition (FER) task. image of xBD dataset

Stable Diffusion Model Compositionality & Category Theory, Singh’s CV Group

  • Exploring the compositionality of StableDiffusion by studying the structure of the latent spaces based on the category theory, and developing loss functions to train the morphisms and functors based on the category theory
  • Creating the testbed for text-image pairs to test for the text compositionality

image of xBD dataset

Spatial-Temporal Change Detection, UW-Madison

  • Developing a contrastive representation learning model for anomaly detection, focusing on detecting and categorizing house, road, and farm damage caused by natural disasters
  • Training on the modified xBD dataset and improving the accuracy of self-supervised change detection model STRCLR by modifying the model to accomplish the anomaly detection downstream task

image of xBD dataset

CARLA Research Project, UW-Madison

  • Customizing the sensors on CAV and roadside units (RSU) using CARLA Python API to collect images and CARLA simulated event data
  • Simulating the operational life cycle of Collaborative Automated Driving System in the CARLA environment

image of CARLA image of CARLA

HuBMAP + HPA: Hacking the Human Body

  • Identify and segment functional tissue units across five human organs
  • Preprocessed the FTUs dataset; developed and fine-tuned Co-Scale Cov-Attentional Image Transformers (CoaT) model to perform segmentation task

image of FTU dataset