AI

DEPICT: Scoring Text-to-Image Alignment by Answer Agreement

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 10:18:24 UTC.

Briefings

DEPICT is a training-free metric for text-to-image alignment that replaces fixed reference answers with expected agreement between image-based and caption-only answers. It achieves an 88% negation accuracy improvement over existing methods. Evaluated on five benchmarks and eleven backbones from three model families, DEPICT surpasses all training-free metrics and outperforms fine-tuned evaluators on two out of three human-correlation benchmarks.

Key points

• DEPICT replaces fixed reference answers with expected agreement between image-based and caption-only answers to improve negation accuracy.
• DEPICT achieves 88% negation accuracy on five benchmarks, surpassing all training-free metrics and outperforming fine-tuned evaluators on two out of three human-correlation benchmarks.

Context

DEPICT addresses core problems in computer vision benchmarking by replacing fixed reference answers with expected agreement between image-based and caption-only answers. This approach increases negation accuracy from 19% to 88% compared to existing approaches.

Limitations

DEPICT evaluation covers five benchmarks and eleven backbones from three model families. Results are based on preprint evidence with no additional background or validation provided.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 19/10/2026, 19:26:11 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeifbcqxc3wcvumauag3qbhryugowdqntfozz7kgilnqmiizbef7cai