Deep-learning algorithm predicts photos’ memorability at “near-human” levels

2015-12-16 · Source: MIT News - Object recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Intermediate, short

Summary

Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed "MemNet," a deep-learning algorithm that predicts image memorability with near-human accuracy. Released on December 16, 2015, this algorithm generates a heat map for any given image, highlighting its most memorable and forgettable regions. The team envisions applications in marketing, education, and personal memory assistance, potentially through an app that subtly modifies photos to enhance or reduce their memorability. As part of this project, CSAIL also published LaMem, the world's largest image-memorability dataset, containing 60,000 annotated images to foster further computer vision research. MemNet utilizes neural networks, a deep-learning technique, to identify patterns in vast datasets without human guidance, performing 30% better than previous algorithms.

Key takeaway

For Computer Vision Engineers developing content creation tools, MemNet offers a novel approach to optimizing visual information. You can integrate this deep-learning technique to predict and manipulate image memorability, enhancing effectiveness in advertising, educational materials, or social media. Consider exploring the LaMem dataset to further refine models or develop new applications focused on human visual memory, potentially improving user engagement and retention.

Key insights

A deep-learning algorithm predicts image memorability with near-human accuracy, identifying key memorable regions.

Principles

• Deep learning can tackle higher-level cognitive processes like human memory.
• Image memorability can be quantified and predicted computationally.

Method

MemNet uses neural networks trained on tens of thousands of images with human-assigned memorability scores to predict how memorable new images will be and generate heat maps of memorable regions.

In practice

• Subtly tweak images to increase or decrease their memorability score.
• Use heat maps to identify most memorable parts of an image.
• Leverage the LaMem dataset for further computer vision research.

Topics

• MemNet Algorithm
• Image Memorability Prediction
• Deep Learning
• Computer Vision
• LaMem Dataset

Best for: AI Scientist, Research Scientist, Computer Vision Engineer, AI Researcher, Machine Learning Engineer, Marketing Professional

Related on AIssential

• Modeling visual memorability assessment with autoencoders reveals characteristics of memorable images — Autoencoder-based models can identify visual features that predict image memorability.
• Non-frontal face recognition using GANs and memristor-based classifiers — Combining lightweight GANs for pose frontalisation with memristor-based neuromorphic recognition enables efficient non-frontal face recognition at the edge.
• Deep Model for Vision — Deep learning overcomes classical computer vision limitations, enabling scalable and adaptable solutions across diverse visual tasks.
• MemDLM: Memory-Enhanced DLM Training — MemDLM uses bi-level optimization and parametric memory to align DLM training with multi-step inference, improving convergence and long-context understanding.
• Beyond Visual Memory: Mechanistic Diagnostics of Latent Visual Reasoning — Gains in latent visual reasoning stem from boundary markers and format, not visual memory encoded in latent slots.
• Vision Hopfield Memory Networks — V-HMN uses hierarchical Hopfield memory and predictive-coding-inspired refinement for data-efficient, interpretable vision backbones.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT News - Object recognition.