I built cephalopod_sentiment_analysis as my submission for the GSoC 2026 entry task, “Sentiment Analysis of Cephalopods.” It is an interpretable, video-based feature-extraction pipeline for cephalopod behavioral analysis using motion and histogram dynamics. My repository listing is also recorded on github profile.
My README presents this repository as my submission for the GSoC 2026 entry task for the project “Sentiment Analysis of Cephalopods.”
I chose simple, interpretable baselines before moving toward complex multi-modal modeling.
My repository is a Jupyter Notebook with 1 star.
My repository started 2026-03 and was last updated 2026-03.
The pipeline handles a short aquatic animal video, reports basic video properties, extracts two visual signals, and plots those signals over time alongside representative frames.
My example video is 10–30 seconds long.
The listed video file is data/octopus_camouflage.mp4.
If the video is not included due to size, I direct people to place their own video in the data/ folder.
I compute motion magnitude using frame differencing; it measures how much the scene changes between frames and indicates activity level.
I compute histogram change using grayscale histogram differences; it captures changes in intensity distribution and can reflect motion, posture, or appearance variations.
- My script reports frame rate (FPS), resolution, total frames, and duration.
- My outputs include Motion vs Time, Histogram Change vs Time, and beginning, middle, and end frames.
My README reports noticeable motion-magnitude spikes around 3–5 seconds, indicating increased activity.
My README reports that histogram change also rises near those intervals, suggesting simultaneous visual variation.
My README says periods with low motion and low histogram change correspond to relatively stable or calm behavior.
I describe higher motion values as possible indications of rapid movement, escape behavior, or active exploration; lower values may correspond to resting or calm states.
I describe histogram changes as potentially reflecting movement, posture changes, or body orientation.
When both features increase together, it may suggest heightened activity or behavioral transitions.
My README cautions that behavioral sentiment cannot be reliably inferred from video alone.
My features do not capture posture or fine-grained body dynamics, and motion magnitude is sensitive to camera movement and noise.
My README notes that histogram change can be affected by lighting variations.
My features are sensitive to background disturbances.
- For future work, I list optical flow for richer motion representation, contour-based analysis for body shape and posture, multi-modal integration of audio and video, and temporal modeling for behavior sequences.
git clone https://github.com/melonmusk20/cephalopod_sentiment_analysis.git
cd cephalopod_sentiment_analysis
pip install -r requirements.txt
python src/extract_features.py
I direct readers to open notebooks/analysis.ipynb to view the exploratory workflow and plots.
My demo video is https://youtu.be/VtCRYbJeNBE.
- My README lists sample outputs at
outputs/feature_plot.png,outputs/sample_frames.png, andoutputs/terminal_output.png.