Unverified50% confidenceFactExact time
Shifting an image by one pixel can cause a CNN model's recognition result to change drastically or even fail completely.
1
Sources
50%
Confidence
Long-term
Relevance
8/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAfter superimposing tiny perturbations that are nearly invisible to the human eye onto an image correctly classified by a neural network, the model output will produce an error with extremely high confidence.77% similarUnverified图生视频的失败几乎都发生在最后一秒，模型会突然snap向参考图导致脸部变形、透视跳变或用白光掩盖生硬过渡74% similarUnverifiedVision Transformer在大规模数据集上超越CNN，但在小数据集上表现不如CNN，因其缺乏图像特有的归纳偏置72% similarUnverifiedResearch by Zhang (2019) demonstrated that standard pooling and downsampling break translation equivariance, and shifting the input by a single pixel can cause non-trivial changes in intermediate feature maps.71% similarUnverified视觉上的剧烈变化未必等同于语义上的重要性，可能导致空镜头触发大量采样而静止画面的关键表情被忽略70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/825956API
curl https://kongchang.com/api/v1/knowledge/claims/825956MCP
get_claim(id=825956)