A Hierarchical Approach for Generating Descriptive Image Paragraphs

[heydenberk]

https://arxiv.org/abs/1611.06607

[nl]

From the title I thought someone had a system that could generate decent paragraph length image descriptions. First I thought 'surely not', then clicked through and saw the authors (Esp. Fei Fei Li) and thought 'oh shit, maybe?', then glanced at the results and realised it is mostly about establishing a baseline system.

We are still 3 years off this, and 5 years wouldn't surprise me. Text is hard.

[pesenti → nl]

I believe you misread the article. It is indeed about a software generating paragraph length description. The article gives a few example outputs. It's not yet human like but getting there.

[nl → pesenti]

Yeah. They are pretty clearly baseline systems, designed to set a benchmark for whatever comes next.