Sentiment Analysis of Cephalopods

Cephalopods display complex behavioral states such as stress, curiosity, aggression, and comfort, but interpreting these states currently depends on expert manual observation and lacks scalable, reproducible tools. This project will build an open source multimodal pipeline that automatically infers and classifies cephalopod behavioral sentiment from video and optional bioacoustic data using computer vision and deep learning. The pipeline will include a dataset ingestion and preprocessing module supporting common video and audio formats, a behavioral feature extractor capturing locomotion, skin pattern dynamics driven by chromatophores, and body posture changes, and a sentiment classification system mapping extracted features to labeled behavioral states. A lightweight deployment API and interactive demo dashboard will make the system accessible to researchers and educators. Key deliverables include dataset ingestion and preprocessing scripts, a multimodal model baseline combining video and optional audio streams, a behavioral feature extraction module for movement and color and posture signals, a sentiment label classifier with training and evaluation scripts, full documentation with reproducible experiments, and a REST API with a demo dashboard for inference. The system will be designed for extensibility across species, datasets, and deployment environments including edge devices.