home.py

import streamlit as st

st.title('batik')
st.subheader('Abstract')
st.markdown('Quantitative analysis of animal behavior represents a burgeoning frontier in neuroscience and ethology. Recent years have witnessed a proliferation of computational methods aimed at identifying behavioral subtypes, or "syllables," from video data. However, while significant advances have been made in behavior segmentation, comparatively few approaches address the interpretation of these behavior syllables, leaving researchers to spend considerable time curating and interpreting the characteristics of the behavioral subtype. Furthermore, most current techniques rely heavily on pose estimation—a prerequisite that, while useful, can introduce limitations concerning generalization in behavioral classification and discovery. Here, we introduce Batik, a system leveraging pre-trained and fine-tuned multimodal transformers to perform end-to-end behavior analysis directly from raw video. Batik excels at supervised behavior annotation, utilizing lightweight models trained on the transformer-extracted feature space to achieve state-of-the-art performance. By integrating a pre-trained vision transformer with a custom fine-tuned language model, Batik not only discovers behavior syllables but also provides expert-level interpretations of mouse behavior, directly from visual data. This comprehensive platform empowers researchers with automated behavior discovery and interpretation, significantly reducing the time burden on experimentalists. Coupled with an intuitive user interface, Batik offers a transformative tool for the next generation of behavioral analysis, showcasing the potential of what is possible with transformer-based language models for behavior.')