Good and bad classification of cooked Idli

This project, titled "Good and Bad Classification of Idli", is designed to develop an image classification system that distinguishes between fresh, properly prepared (good) idlis and spoiled, fungus-affected (bad) idlis.
The dataset consists of approximately 2000 images, evenly distributed between good and bad samples (1000 good and 1000 bad).
All images were captured using the Infinix GT 20 Pro smartphone, which provides high-resolution image quality suitable for machine learning and computer vision applications. The idlis were photographed against a white chart paper background under natural daylight conditions to ensure uniformity, clarity, and accurate feature capture.
Device Specifications (Data Acquisition Device)
All images were captured using the Infinix GT 20 Pro smartphone. The important specifications of the device are:
Rear Camera: 108 MP (Primary Wide Sensor)
Image Resolution: High-resolution image capture suitable for detailed texture analysis
Battery Capacity: 5000 mAh
The 108 MP rear camera was used for capturing the dataset images, ensuring high clarity, sharp texture details, and accurate color representation necessary for detecting fungal growth and discoloration in idlis.
Dataset Composition
Good Samples (Fresh Idli)
The dataset includes approximately 1000 images of good-quality idlis. These images show idlis with:
Soft and fluffy texture
Proper white color
Smooth and even surface
Well-formed round shape
No visible discoloration
No fungal growth
These samples represent the positive class and help train the model to recognize fresh and properly prepared idli conditions.
Bad Samples (Fungus-Affected Idli)
The dataset also contains approximately 1000 images of bad-quality idlis. These idlis may exhibit:
Discoloration (yellowish or dull appearance)
Visible fungal growth (green/black spots)
Surface texture degradation
Signs of over-fermentation
Poor overall physical condition
These images represent the negative class, enabling the model to identify spoiled or contaminated idlis accurately.
Data Collection Setup
All images were captured using the Infinix GT 20 Pro smartphone under controlled conditions.
A white chart paper background was used intentionally to:
Provide a uniform and clean background
Enhance contrast between the idli and background
Reduce noise and unwanted visual distractions
Improve visibility of texture, surface features, and fungal growth
Images were taken under natural daylight conditions, ensuring consistent lighting and accurate representation of color and texture.
Each image contains three idlis, photographed from different angles to introduce variation in orientation and improve model robustness.
Image Characteristics
The dataset includes variations in:
Idli size
Surface texture
Color intensity
Angle of capture
Health condition (fresh vs fungus-affected)
Data Annotation
Each image is carefully labeled as either:
"Good" (Fresh Idli)
"Bad" (Fungus-Affected Idli).