Intrinsic Image Popularity Assessment with PyTorch Implementation

Intrinsic Image Popularity is a dataset and PyTorch codebase for assessing the inherent popularity of images, as described in the associated research paper. The repository, created by author dingkeyan93, was last updated on November 25, 2023. It contains code for training and evaluating models on image popularity metrics, likely derived from social media platforms like Instagram.

Use Cases

• Train a neural network to predict image popularity scores from visual features extracted by a CNN.
• Benchmark different model architectures (e.g., CNNs, transformers) on the provided image popularity labels.
• Analyze the correlation between intrinsic image attributes and social media engagement metrics.
• Fine-tune pre-trained image encoders for the specific task of popularity assessment.

Strengths

• Provides a complete PyTorch implementation for reproducibility of the research paper's results.
• Includes code for model training and evaluation on a defined image popularity task.

Limitations

• The underlying dataset size, number of images, and specific column features are not documented.
• Potential geographic or platform bias exists if data is sourced exclusively from Instagram.

Provenance

Source

GitHub repository by dingkeyan93, associated with an academic paper.

Collection Method

Methodology from the paper 'Intrinsic Image Popularity Assessment'; specific data collection method is not detailed.

Freshness

Last updated in November 2023.