Project evidence / Early work
Emotion-Aware Food Recommender
The checked-in source implements grayscale conversion, histogram equalization, Gaussian smoothing, original-versus-preprocessed inference, and a fixed emotion-to-food mapping. The repository does not contain a recorded evaluation proving that preprocessing improves accuracy or robustness.
- Status
- Early work
- Evidence
- verified repository
- Last reviewed
- 2026-09-08
Overview
An early Streamlit image-processing prototype that captures a visitor-initiated webcam image, preprocesses it, applies a pretrained FER model, and maps the detected emotion to food suggestions.
Early image-processing prototype; no model-training or measured quality claim is made here.
Problem
Explore whether a simple preprocessing path can make webcam emotion input easier to inspect before mapping results to recommendations.
Implementation
A Streamlit app decodes a captured image, applies OpenCV preprocessing, runs the same pretrained FER detector on original and processed images, and displays both outputs.
- Visitor-initiated webcam capture
- Grayscale, histogram-equalization, and Gaussian-blur preprocessing
- Pretrained facial-emotion inference
- Fixed emotion-to-food recommendation mapping
Security and reliability
- The camera input is explicitly visitor initiated within the Streamlit interface.
Limitations
- No recorded comparison supports improved-accuracy or illumination-robustness claims.
- Dependencies use open version ranges and no lockfile is present.
- No tests, CI, license, FER2013 attribution detail, or webcam privacy documentation are present.
- One caught detector exception is displayed directly to the visitor.