awgs-vision-v0.3

AWGS Vision v0.3

<p align="center"> <img src="assets/AWGS-vision.png" alt="AWGS Vision v0.3"/> </p> <p align="center"> <img src="https://img.shields.io/badge/status-experimental-orange" alt="Status: Experimental"/> <img src="https://img.shields.io/badge/base_model-Gemma%204%20E2B-blueviolet" alt="Base Model: Gemma 4 E2B"/> <img src="https://img.shields.io/badge/training-Unsloth-yellow" alt="Training: Unsloth"/> <img src="https://img.shields.io/badge/task-Image%20Safety%20Classification-red" alt="Task: Image Safety Classification"/> <img src="https://img.shields.io/badge/version-v0.3-blue" alt="Version: v0.3"/> </p>

AWGS Vision v0.3 is an experimental multimodal vision model designed for image safety and content classification. It is fine-tuned from Google's Gemma 4 E2B and trained to analyze images while returning structured safety-related predictions.

AWGS Vision v0.3 builds upon the previous AWGS Vision releases and focuses on improving robustness across diverse visual conditions.

The model was trained on a custom multimodal instruction dataset prepared for AWGS Vision.

---

Table of Contents

• Overview
• The AWGS Vision Family
• What's New
• Model Details
• Capabilities
• Limitations
• Intended Use
• Architecture & Training
• Output Format
• Evaluation
• Risks & Considerations
• Community & Support
• License
• Acknowledgements
• About AWGS
• Image Credits

---

Overview

AWGS Vision v0.3 is an experimental multimodal vision model designed for image safety and content classification.

The model receives an image together with an instruction and generates a textual response containing safety-related predictions.

AWGS Vision v0.3 was fine-tuned from Gemma 4 E2B using the Unsloth ecosystem.

This release focuses on improving the model's classification performance and robustness across a variety of visual conditions.

---

The AWGS Vision Family

AWGS Vision is an experimental model series developed under ArkAiLabs, focused on practical and efficient multimodal models for image analysis and safety-oriented classification.

The project explores the use of vision-language foundation models for specialized image-processing workloads, including image safety classification, NSFW and non-NSFW classification, experimental content moderation, multimodal model benchmarking, and lightweight image-analysis pipelines.

AWGS Vision v0.3 builds upon previous releases with continued improvements to classification performance, robustness, and generalization.

Previous Releases

• AWGS Vision v0.1 — Initial experimental release.
• AWGS Vision v0.2 — Introduced additional image variations in training.
• AWGS Vision v0.3 — Further expands training data diversity.

Future releases may improve classification accuracy, robustness, output consistency, confidence calibration, and inference efficiency.

---

What's New

AWGS Vision v0.3

AWGS Vision v0.3 continues the development of the AWGS Vision model family with improvements to image safety classification, robustness, and generalization across diverse visual inputs.

This release focuses on improving model performance while maintaining the lightweight and practical design of the AWGS Vision family.

Specific dataset composition and training augmentation details are not publicly disclosed.

---

Model Details

---

Capabilities

AWGS Vision v0.3 is designed to:

• Analyze images for safety-related content
• Classify images into broad safety categories
• Identify potentially safe, suggestive, explicit, or violent content
• Produce structured classification responses
• Provide confidence-like scores for supported categories
• Process image inputs together with natural-language instructions
• Support prototype image-moderation systems
• Handle diverse visual conditions in input images

Example output:

{
  "safe": 0.96,
  "suggestive": 0.12,
  "explicit": 0.01,
  "violence": 0.03,
  "result": "SAFE"
}

Another possible output:

{
  "safe": 0.04,
  "suggestive": 0.71,
  "explicit": 0.94,
  "violence": 0.08,
  "result": "NSFW"
}

The exact output format is not guaranteed for every generation. Applications should validate, sanitize, and parse the model response before using it programmatically.

---

Limitations

AWGS Vision v0.3 is an experimental model and has important limitations.

Benchmark Limitations

Benchmark results should be interpreted according to the evaluation dataset and methodology used.

A benchmark result does not automatically represent real-world accuracy or generalization to unseen data.

Future evaluations should use larger, diverse, independently held-out datasets.

Image Variation Limitations

Although v0.3 was trained using diverse image variations, performance on arbitrary or extreme visual changes is not guaranteed.

Different levels of compression, resizing, cropping, noise, or other transformations may produce different results.

Background Sensitivity

Classification performance may vary depending on image background and overall visual composition.

Changes in lighting, scenery, color distribution, shadows, image noise, or surrounding objects may influence the model's prediction.

Output Consistency

The model may occasionally generate different JSON structures from the expected format.

For example:

{
  "label": "safe"
}

or:

{
  "is_nsfw": false
}

Applications should not assume that every response will exactly match the AWGS output schema.

Use structured parsing, schema validation, fallback handling, and error logging in production-like prototypes.

Confidence Scores

The numerical values generated by the model represent confidence-like scores.

They should not be interpreted as calibrated probabilities.

A score of 0.98 does not necessarily mean that the model has a statistically calibrated 98% probability of being correct.

Dataset Limitations

The training dataset may contain ambiguous examples, labeling errors, dataset bias, limited visual diversity, uneven representation of content types, or images that do not reflect real-world distributions.

Generalization

Performance may vary depending on image type, resolution, quality, cropping, framing, lighting, background composition, visual ambiguity, compression, prompt wording, and differences between training and deployment data.

Production Readiness

AWGS Vision v0.3 is not production-ready.

It should not be used as the sole decision-making component in automated safety-critical systems or high-consequence content moderation systems.

Human review and additional validation are recommended for important decisions.

---

Intended Use

Supported Use Cases

AWGS Vision v0.3 is intended for:

• AI and multimodal research
• Image classification experiments
• Prototype moderation systems
• Educational experimentation
• Benchmarking vision-language models
• Development of image-processing pipelines
• Research into lightweight vision inference
• Testing image safety workflows
• Exploring structured outputs from multimodal models
• Researching model robustness against visual variations

Out-of-Scope Use Cases

The model is not intended for:

• Fully autonomous content moderation without additional safeguards
• Safety-critical decision-making
• High-stakes classification
• Applications requiring guaranteed accuracy
• Systems where incorrect classifications could cause significant harm
• Automatic enforcement actions without human review or secondary validation
• Medical, legal, or law-enforcement classification

Developers should combine model predictions with policy rules, validation checks, uncertainty handling, and human review where appropriate.

---

Architecture & Training

AWGS Vision v0.3 is based on Google Gemma 4 E2B, a multimodal vision-language model designed to process image and text inputs.

Training Configuration

AWGS Vision v0.3 was fine-tuned from the Unsloth-provided unsloth/gemma-4-E2B model, which is based on Google's google/gemma-4-E2B.

Unsloth was used for the fine-tuning workflow and training optimization.

The resulting model was trained on a custom multimodal instruction dataset prepared for AWGS Vision.

The v0.3 dataset expands the visual diversity of the training data through additional image variations.

---

Output Format

The intended AWGS output format is:

{
  "safe": 0.00,
  "suggestive": 0.00,
  "explicit": 0.00,
  "violence": 0.00,
  "result": "SAFE"
}

or:

{
  "safe": 0.00,
  "suggestive": 0.00,
  "explicit": 0.00,
  "violence": 0.00,
  "result": "NSFW"
}

The numerical values represent model-generated confidence-like scores and should not be interpreted as calibrated probabilities.

Because the model generates text, responses may occasionally contain additional explanation, invalid JSON, missing fields, alternative field names, or incomplete output.

A safer integration should extract the generated response, parse the expected JSON object, validate required fields and value ranges, apply fallback behavior when parsing fails, and route uncertain or invalid cases for additional review.

---

Evaluation

AWGS Vision v0.3 should be evaluated using both regression testing and independent held-out evaluation.

Internal Regression Benchmark

Previous Benchmark

AWGS Vision v0.2 achieved 98% accuracy on the project's 100-image internal regression benchmark.

AWGS Vision v0.1 achieved 44% accuracy on the same benchmark.

These results were intended for release-to-release comparison and should not be interpreted as independent estimates of real-world performance.

Future releases should be evaluated using larger, more diverse, independently held-out datasets.

Additional metrics such as precision, recall, F1 score, confusion matrices, per-category performance, and calibration error would provide a more complete evaluation.

---

Risks & Considerations

Developers integrating AWGS Vision v0.3 should account for:

• False positives
• False negatives
• Background sensitivity
• Sensitivity to visual variations and transformations
• Inconsistent output formatting
• Dataset bias
• Ambiguous image classifications
• Confidence-score calibration limitations
• Distribution shifts between benchmark and real-world images
• Performance degradation on unseen image types
• Incorrect predictions caused by image quality or preprocessing
• Prompt sensitivity

The model should not be the sole moderation or safety mechanism in an application.

A robust deployment should combine model predictions with additional validation, explicit moderation policies, rule-based checks, threshold tuning, logging and monitoring, human review for uncertain cases, periodic evaluation on representative deployment data, and safe handling of malformed model outputs.

---

Community & Support

AWGS Vision is an experimental project developed under ArkAiLabs.

Community feedback, testing, benchmark contributions, and research collaboration are welcome as the project develops.

For information about previous releases, visit the AWGS Vision model family on Hugging Face.

---

License

AWGS Vision v0.3 is based on Google Gemma 4 E2B.

The model is released under the Apache 2.0 license, subject to the applicable terms and conditions of the underlying Gemma model.

---

Acknowledgements

AWGS Vision v0.3 builds upon Gemma 4 E2B by Google DeepMind and the Unsloth-provided unsloth/gemma-4-E2B model.

We thank the Unsloth team for their fine-tuning tools and optimized model implementations that supported the development of AWGS Vision.

We also thank the open-source AI and machine-learning community for the tools, research, and resources that make projects like AWGS Vision possible.

---

About AWGS

AWGS Vision is an experimental vision model project under ArkAiLabs, focused on developing practical and efficient multimodal AI systems for image analysis and safety-oriented classification.

AWGS Vision v0.3 represents the third public release of the project and continues the development of the AWGS Vision model family through expanded data diversity and robustness experimentation.

---

Image Credits

The background image used in the AWGS Vision v0.3 model artwork was photographed by Eberhard Grossgasteiger and is sourced from Unsplash.

• Photographer: Eberhard Grossgasteiger
• Image: A house sitting on top of a hill covered in fog
• Source: Unsplash
• License: Unsplash License