Unverified50% confidenceFactExact time
buffalo_l在MS1MV3数据集（约300万张图像）上训练，LFW准确率接近99.8%
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
IMGNet: A New Face Verification Approach Challenging Cosine Similarity with Sign Pattern Matching
redditr/computervision7/9/2026
Related Claims
UnverifiedPathway reported that a BDH-based system achieved a 97.4% Top-1 accuracy rate on approximately 250,000 Sudoku Extreme puzzles without using CoT or solution backtracking.59% similarUnverified一个rank=4的LoRA文件对于SD1.5模型约只有3-6MB，LoRA设计使训练参数量减少95%以上58% similarUnverifiedBullet 在 SWE-bench Verified 基准测试中取得 95.8% 的成绩，位列前三，平均每个任务耗时 119 秒56% similarUnverifiedHy3 adopts a Mixture of Experts (MoE) architecture with a total parameter count of 295 billion (295B), activating only 21 billion (21B) parameters per inference.56% similarUnverifiedLFM2.5在多项基准测试中展现出与体量高达其4倍模型相当的竞争力56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/484238API
curl https://kongchang.com/api/v1/knowledge/claims/484238MCP
get_claim(id=484238)