> ML_BENCHMARK // IMAGENET-1K-TOP1-ACCURACY_v1.0
ImageNet-1K Top-1 Accuracy (ILSVRC 2012 Validation)
Computer Vision Classification · task-image-classification · near-saturation
Computer Vision Classificationnear-saturationhigh-vram-dependent
Evaluation Protocol
50,000 validation images evaluated on center crop or multi-crop without test-time augmentation bells and whistles.
Baseline & Metrics
Canonical Baseline:AlexNet (2012): 57.1% | ResNet-50: 76.0% | ViT-H/14: 88.5% | CoAtNet-7: 90.88%
Evaluated Metrics:
Top-1 Accuracy (%)Top-5 Accuracy (%)
Contamination & Leakage Risks
Widely benchmarked for over 12 years with minimal test leak; near saturation on human agreement upper bound (~92%).
Reproducibility Concerns
Image crop size (224x224 vs 384x384 vs 512x512) and interpolation method (bicubic vs bilinear) shift scores by 1-2%.
CONNECTED DATASETImageNet ILSVRC 2012 (ImageNet-1K)Computer Vision & Object Recognition · 1,281,167 training images across 1,000 synsets
View Dataset Spec