Unverified50% confidenceBenchmarkExact time
When trained from scratch on ImageNet-1K, SALU reaches 80.16% Top-1 with ResNet-50 and 80.28% Top-1 with ViT-B/16, using only BS=128.
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn CIFAR-100 (ResNet-18), SALU achieves 83.42% Top-1 at BS=512 and still retains 76.23% Top-1 at BS=1, whereas standard BatchNorm collapses at BS=1.72% similarUnverifiedIn the MicroLens-100k evaluation, JEV-27B-VL using only cover-image scoring reached 0.727 AUC and 0.590 HR@5 among 20 candidates, while collaborative filtering trained on the history of 59,045 users achieved 0.728 AUC and 0.490 HR@5.64% similarUnverifiedThe V1 approach achieved a self-tested top-1 accuracy of 93%, but scored 0/137 top-5 accuracy on a benchmark set of 137 real photographs.63% similarUnverifiedB7 ranks first on ImageNet, whereas in this OCT four-class task B7 shows only a secondary accuracy rebound and does not surpass B1.63% similarUnverifiedWhen DINOv3 ViT-H+/16 is pruned to 50% sparsity, removing 427 million parameters, ImageNet-1k linear probe accuracy drops by only 0.8%, without weight correction or retraining.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1151070API
curl https://kongchang.com/api/v1/knowledge/claims/1151070MCP
get_claim(id=1151070)