vhchuong1997 / salicon-evaluation

evaluation API for salicon : a counterpart of saliency prediction as https://github.com/tylin/coco-caption

SALICON Saliency Evaluation

Saliency evaluation codes for SALICON dataset.

Requirements

• python 2.7

Files

./

• saliconEvalDemo.py (demo script)
• image2json.py (converts a folder of saliency maps to result JSON)

./annotation

• fixations_val2014.json (SALICON 2014 validation set)
• Visit SALICON download page for more details.

./results

• fixations_val2014_fake_results.json (an example of fake results for running demo)
• Visit SALICON format page for more details.

./saliconeval: The folder where all evaluation codes are stored.

• evals.py: The file includes SALICONEval class that can be used to evaluate results on SALICON.
• auc: AUC evalutation code
• sauc: Shuffled AUC evaluation code
• nss: NSS evaluation code
• cc: CC evaluation code

References

• SALICON: Saliency in Context
• Microsoft COCO Captions: Data Collection and Evaluation Server
• Metrics code (AUC, Shuffled AUC, NSS and CC) are migrated from the MATLAB code of the [MIT Saliency Benchmark] (http://saliency.mit.edu/)

Developers

• Ming Jiang
• Shane Huang

Acknowledgement

• COCO Consortium