Unverified50% confidenceFactExact time
During image downsampling, key features of a target that occupies a very small proportion of the image may be lost, causing the model to fail to see it clearly.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe quality of a visual localization system's predictions is highly dependent on the number of usable geographic cues present in the image, and information-sparse images can put the model in a difficult position.79% similarUnverified在《威利在哪里》中威利通常只占据整幅图像面积的极小一部分（小于0.1%），模型在有限视觉token预算下很难看清目标76% similarUnverifiedWhen lacking explicit functional priority constraints, models tend to maximize visual salience rather than usability75% similarUnverified超过API图像限制的图像会被静默缩放，导致模型基于降质图像进行点击预测，造成坐标偏移74% similarUnverified逐图 min/max 归一化会抹掉图像间的绝对强度差异，对依赖光谱绝对值区分地物的遥感分割任务可能丢失关键信息74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/909703API
curl https://kongchang.com/api/v1/knowledge/claims/909703MCP
get_claim(id=909703)