AI CASE STUDY: DEFORMATION (NSFW) CLASSIFICATION
Project DESCRIPTION:
§ Automated deformation (NSFW) classification distinguishes between images of normal individuals and those depicting sensitive or inappropriate content in the workplace, streamlining the process by replacing manual efforts with automated image processing. It effectively identifies images of healthy individuals versus those showing inappropriate characteristics.
§ This project robustly identifies, classifies, and evaluates diverse input images to determine their suitability for the workplace environment. Developed and tested on various solutions, and trained on a large dataset, this project achieves optimized recognition capability and high accuracy.
BUSINESS CHALLENGES:
Content Moderation on Digital Platforms
Inappropriate Content Affecting Work Environment
Legal Risks and Regulatory Compliance
TITAN SOLUTION:
Solution 1: Keras
§ Developed in Python using prominent Keras CNN models such as ResNet50 and MobileNetV2, the project employs a training dataset gathered from various sources like social media, magazines, and newspapers. The input dataset is meticulously filtered to reduce noise and undergoes detailed classification and labeling for each data entry. Subsequently, it undergoes multiple layers of image processing, including data augmentation, normalization, and feature extraction crucial for effective model learning.
§ These steps are followed by rigorous model training experiments aimed at achieving optimal accuracy for practical deployment.
Solution 2: AI Platform
§ The input dataset, sourced from social media, magazines, and newspapers, is used for image-based AI training. Leveraging the AI platform’s robustness, the Deformation (NSFW) classification model is trained on this dataset efficiently, ensuring high accuracy and performance on diverse real-world datasets.
§ Diverse images, categorized into at least two classes with a minimum of 10 images per class, are uploaded to the system. AutoML automates the model training, enabling quick testing and evaluation of accuracy. Images must be under 6MB and in JPG, PNG, BMP, or WEBP format.
BUSINESS IMPACT :
# Positive:
§ Fostering a healthy and positive work environment
§ Efficient content moderation eliminates manual review efforts
§ Strict compliance with regulations and legal standards
§ Enhanced User Experience
§ Improved User Safety
# Negative:
§ Accuracy limitations in machine learning compared to human precision.
§ Using Keras requires investing in powerful CPUs and GPUs for effective model training, while an AI platform incurs monthly costs for project expansion.
