EduScale AI Model Card

Model Details

EduScale AI provides two TensorFlow Lite super-resolution models for educational image enhancement:

Model	Scale	File
EduScale SPAN x2	2x	`models/eduscale_span_x2.tflite`
EduScale SPAN x3	3x	`models/eduscale_span_x3.tflite`

The models are intended for mobile and edge use cases where low-resolution slide or study-material images need to be enhanced before reading or OCR.

Intended Use

Educational slide image enhancement.
OCR preprocessing for screenshots, slides, and study materials.
Mobile deployment using TensorFlow Lite.
Research, demonstration, and portfolio evaluation.

Data

Project data was gathered from educational presentation material and creator-made slide/image samples. The public source identified for part of the collected data is Kaggle's Dataset for ppt by Manisha717, described as PPT files for testing across varied topics.

Because the Kaggle page lists the dataset license as unknown, this model repository does not redistribute the original PPT files or source images.

Out-of-Scope Use

Medical, legal, or safety-critical document interpretation.
Recovering text that is not visually supported by the original image.
Treating OCR output as verified ground truth without review.

Benchmark Results

The following values are averages from the detailed benchmark CSV files:

Model	Scale	PSNR	SSIM	OCR Confidence	CER	Runtime	Device
`eduscale_span_x2`	2x	29.46	0.9786	91.23	0.0736	573 ms	Realme Note 50
`eduscale_span_x3`	3x	26.60	0.9642	89.17	0.1980	288 ms	Realme Note 50

Detailed results are available in:

benchmarks/benchmark_x2.csv
benchmarks/benchmark_x3.csv
benchmarks/benchmark_summary.csv

Metrics

PSNR: image reconstruction quality; higher is better.
SSIM: structural similarity; higher is better.
OCR confidence: OCR model confidence after super-resolution; higher is better.
CER: character error rate after super-resolution; lower is better.
runtime_ms: average measured inference runtime on the listed device.

Limitations

Performance can vary by camera quality, compression artifacts, text size, font, language, and lighting.
Super-resolution may sharpen artifacts along with useful text features.
OCR gains are not guaranteed for every image.
The benchmark set is education-focused and may not represent all document or natural-image domains.
Source data includes a public Kaggle PPT dataset with unknown license status, so users should review upstream terms before reusing the original PPT data directly.

Ethical Considerations

Enhanced images should not be used to fabricate or overstate document accuracy. When used for OCR or study workflows, extracted text should be verified against the original source when correctness matters.

License

Apache 2.0.