Unverified50% confidenceFactExact time
ScaleReach uses face tracking for 9:16 vertical reframing.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/1/2026
First Seen
Valid until: 1/30/2027
Sources
Related Entities
Related Claims
Unverified端到端可微的人脸相似度链路包括人脸检测（RetinaFace/MTCNN）、人脸对齐（Spatial Transformer Network）、特征提取（ArcFace/FaceNet/AdaFace，输出512维或128维嵌入）和余弦相似度计算62% similarUnverifiedThis image localization uses a traditional computer vision localization algorithm rather than a deep learning model, locating the position by calculating the splice line position, traversing horizontal offsets, and taking an approximately 12px strip region on the splice line to compare the continuity of the upper and lower images.60% similarUnverifiedWhen a dataset contains a large number of half-body or full-body shots, supplementing it with face-cropped images allows the identity-critical features to occupy a larger proportion of the image, thereby improving the capture of facial details.59% similarUnverifiedUsing detected keypoints to rotate and scale a face into a standard pose (face alignment) before extracting feature vectors can greatly improve feature stability.59% similarUnverifiedRestoring facial micro-expressions often requires additionally introducing facial landmark detection or dedicated facial reenactment models.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1123503API
curl https://kongchang.com/api/v1/knowledge/claims/1123503MCP
get_claim(id=1123503)