Clinton
ANCHOR: LLM-driven News Subject Conditioning for Text-to-Image Synthesis
Ramakrishnan, Aashish Anantha, Huang, Sharon X., Lee, Dongwon
Text-to-Image (T2I) Synthesis has made tremendous strides in enhancing synthesized image quality, but current datasets evaluate model performance only on descriptive, instruction-based prompts. Real-world news image captions take a more pragmatic approach, providing high-level situational and Named-Entity (NE) information and limited physical object descriptions, making them abstractive. To evaluate the ability of T2I models to capture intended subjects from news captions, we introduce the Abstractive News Captions with High-level cOntext Representation (ANCHOR) dataset, containing 70K+ samples sourced from 5 different news media organizations. With Large Language Models (LLM) achieving success in language and commonsense reasoning tasks, we explore the ability of different LLMs to identify and understand key subjects from abstractive captions. Our proposed method Subject-Aware Finetuning (SAFE), selects and enhances the representation of key subjects in synthesized images by leveraging LLM-generated subject weights. It also adapts to the domain distribution of news images and captions through custom Domain Fine-tuning, outperforming current T2I baselines on ANCHOR. By launching the ANCHOR dataset, we hope to motivate research in furthering the Natural Language Understanding (NLU) capabilities of T2I models.
- North America > United States > Utah > Salt Lake County > Salt Lake City (0.04)
- North America > Dominican Republic (0.04)
- Oceania > Australia > Victoria > Melbourne (0.04)
- (5 more...)
- Government (0.67)
- Media > News (0.66)
A Few Thoughts on the Unthinkable
Jump to an update: • Waiting for Hillary on a gray morning • A few thoughts on the unthinkable • Palin speaks at the Trump party • Trump's path • Settling in for the night at Hillary Clinton's party • If Trump wins, he would likely also control all three branches of government • A new electoral map is upending the old one • The part of the night when Democrats start to freak out • Marco Rubio, again • The exit polls show a breakdown in demographics that is entirely predictable • A shooting near a polling place in Los Angeles • Early exit polls: No evidence Comey made a difference • Is the South still the conservative heartland? Clinton's motorcade arrived soon after. At campaign events and at her party last night, Clinton was permanently inside a huge bubble of safe space guarded by the Secret Service. At today's event, they were nowhere to be seen. Clinton arrived in a small caravan that stopped in a busy street. The only visible protection was provided by a handful of New York cops who hadn't received notice she was coming just then and halfheartedly tried to convince a crowd to move backward. Soon, Clinton's staff and the crowd and a few people who happened to have been walking down the street were mashed together for a panicky moment. A bicyclist, nearby, screamed, "Get out of the way, you fucking morons."--A. The executive branch of the United States government has grown in its power over the past eight years. After 9/11, George W. Bush built an aggressive national-security apparatus that Barack Obama only partially reined in. To cite just one of the powers that Commander-in-Chief Donald J. Trump would acquire, the American President has grown comfortable with killing alleged terrorists remotely with unmanned vehicles. Congress has done little in the way of oversight of this program, and it is just one of the many new powers Trump could inherit. Similarly, Congress has shown no interest in rewriting the overly broad war authorizations that Bush and Obama used to wage campaigns across the Middle East and Africa. As Congress and the White House became unable to pass legislation, Obama also pushed the boundaries with respect to the use of executive orders. These can be rescinded on day one of a Trump Presidency, but, just as important, Trump will undoubtedly push the boundaries of executive orders beyond what Obama did.
- North America > United States > California > Los Angeles County > Los Angeles (0.24)
- North America > United States > New York (0.24)
- Europe > Middle East (0.24)
- (29 more...)