Unverified50% confidenceFactExact time
Slot Attention was proposed by Locatello et al. in 2020, iteratively assigning image features to a fixed number of slot vectors through competitive attention.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe core insight of Prompt-to-Prompt is that the attention maps of the cross-attention layers in diffusion models directly control the spatial layout of the image.68% similarUnverified原始 Self-Attention 的计算量随序列长度呈平方级增长，Anthropic 通过优化位置编码（如 ALiBi）等技术手段拓展了可用上下文68% similarUnverifiedExisting mainstream sparse attention methods filter interactions by retaining entries with larger values or high-quality regions in the attention matrix.67% similarUnverified自注意力使任意两个位置之间的信息交互只需一步计算，但显存占用随序列长度平方增长66% similarUnverified在密集场景视觉搜索中，可采用先概览全图识别候选区域、再对候选区域裁剪放大（crop-and-zoom）逐步验证的主动视觉探索策略65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/990673API
curl https://kongchang.com/api/v1/knowledge/claims/990673MCP
get_claim(id=990673)