Deep Visual-Semantic Alignments for Generating Image Descriptions Haotian Xu
Roadmap Overview Technical Approach Evaluation
Overview To generate dense and free-form descriptions of images
Overview Contributions 1) Infer region-word alignments 2) Generate model of images descriptions 3) Generate region-level descriptions
Technical Approach Inferring Alignments
Technical Approach
Technical Approach Infer word alignment
Technical Approach Generate descriptions
Technical Approach Generate region-level descriptions
Evaluation
Evaluation -- Alignment
Evaluation -- Description
Evaluation –- Region Level
Thanks!