model_id,model_name,metric,fp32,quantized,delta,published_comparison
AD01,MLPerf Tiny Deep AutoEncoder (DCASE 2020 ToyCar),AUC / pAUC (max_fpr=0.1),0.876001 / 0.764121,0.840250 / 0.720049,-0.035750 / -0.044071,공개 FP32 AUC/pAUC 0.876001/0.764121; 측정값 동일
LM04,BERT-tiny RAID text detector ONNX,AUROC / TPR@FPR 5% / 1% (RAID-extra OOD),0.667078 / 0.318983 / 0.233873,0.668862 / 0.322615 / 0.235853,+0.001785 / +0.003632 / +0.001981,공개 수치 없음.
OD06,EfficientDet-Lite0 int8/float32 v1,COCO bbox mAP (AP@[0.50:0.95]),24.8751%,24.2822%,-0.5929 pp,공개 수치 없음.
OD07,EfficientDet-Lite2 int8/float32 v1,COCO bbox mAP (AP@[0.50:0.95]),31.8594%,31.4191%,-0.4403 pp,공개 수치 없음.
SG06,DeepLabV3-MobileNetV2 width 1.0 Pascal train_aug,mIoU,75.6398%,74.1290%,-1.5108 pp,공개 FP32 mIoU 75.32%; 측정 75.6398% (+0.3198 pp)
SG07,DeepLabV3-MobileNetV2 width 0.5 Pascal train_aug,mIoU,70.6473%,69.6191%,-1.0282 pp,공개 FP32 mIoU 70.19%; 측정 70.6473% (+0.4573 pp)
SG08,EdgeTPU-DeepLab-slim width 0.75 Cityscapes,mIoU (CamVid cross-dataset),50.6498%,51.1600%,+0.5102 pp,공개 수치 없음.
SP01,MLPerf Tiny DS-CNN KWS reference,Top-1 accuracy,91.86%,91.66%,-0.2045 pp,"공개 시험 정확도 약 92%; 측정 FP32 91.86%, 양자화 91.66% (근접)"
SP02,MLPerf Tiny streaming wakeword 1D DS-CNN,FP / FN (1 s),5 / 6,4 / 6,-1 / +0,공개 수치 없음.
SP08,TensorFlow Lite Micro micro_speech tiny_conv,Top-1 accuracy (yes/no subset),94.05%,94.05%,+0.0000 pp,공개 수치 없음.
SP09,Arm ML-Zoo clustered DS-CNN Large INT8,Top-1 accuracy,95.06%,94.70%,-0.3590 pp,공개 수치 없음.
VC01,MLPerf Tiny Visual Wake Words MobileNetV1 0.25,Top-1 accuracy,85.10%,85.60%,+0.5000 pp,공개 수치 없음.
VC02,MLPerf Tiny CIFAR-10 ResNet8,Top-1 accuracy,87.00%,87.00%,+0.0000 pp,공개 수치 없음.
VC03,TensorFlow MobileNetV1 alpha=0.25 224,Top-1 / Top-5 accuracy,49.80% / 74.20%,48.00% / 72.80%,-1.8000 pp / -1.4000 pp,공개 수치 없음.
VC04,TensorFlow MobileNetV1 alpha=0.5 224,Top-1 / Top-5 accuracy,63.30% / 84.90%,60.70% / 83.20%,-2.6000 pp / -1.7000 pp,공개 수치 없음.
VC05,ST MobileNetV2 alpha=0.35 224,Top-1 accuracy,58.13%,56.77%,-1.3600 pp,공개 수치 없음.
VC06,ST MobileNetV2 alpha=0.50 PyTorch 224 QDQ,Top-1 accuracy,66.20%,65.31%,-0.8900 pp,공개 수치 없음.
VC09,ONNX Model Zoo SqueezeNet 1.0 INT8,Top-1 / Top-5 accuracy,56.85% / 79.87%,56.48% / 79.76%,-0.3700 pp / -0.1100 pp,공개 수치 없음.
VC11,Google MediaPipe EfficientNet-Lite0 224 INT8,Top-1 accuracy,75.10%,74.40%,-0.7000 pp,공개 수치 없음.
VC12,ONNX Model Zoo MobileNetV2 1.0 INT8,Top-1 / Top-5 accuracy,69.48% / 89.26%,68.30% / 88.44%,-1.1800 pp / -0.8200 pp,공개 수치 없음.
VC13,ONNX Model Zoo ShuffleNetV2 x1.0 INT8,Top-1 / Top-5 error,33.65% / 13.43%,33.85% / 13.66%,+0.2000 pp / +0.2300 pp,공개 수치 없음.