Files
openvino/docs/benchmarks/performance_int8_vs_fp32.md
2022-12-21 18:05:03 +01:00

2.1 KiB

Model Accuracy and Performance for INT8 and FP32

The following table presents the absolute accuracy drop calculated as the accuracy difference between FP32 and INT8 representations of a model on two platforms

  • A - Intel® Core™ i9-9000K (AVX2)
  • B - Intel® Xeon® 6338, (VNNI)
  • C - Intel® Flex-170

@sphinxdirective .. list-table:: Model Accuracy :header-rows: 1

    • OpenVINO™ Model name
    • dataset
    • Metric Name
    • A
    • B
    • C
    • bert-base-cased
    • SST-2_bert_cased_padded
    • accuracy
    • 0.11%
    • 1.15%
    • 0.57%
    • bert-large-uncased-whole-word-masking-squad-0001
    • SQUAD_v1_1_bert_msl384_mql64_ds128_lowercase
    • F1
    • 0.51%
    • 0.55%
    • 0.68%
    • deeplabv3
    • VOC2012_segm
    • mean_iou
    • 0.44%
    • 0.06%
    • 0.04%
    • densenet-121
    • ImageNet2012
    • accuracy @ top1
    • 0.31%
    • 0.32%
    • 0.30%
    • efficientdet-d0
    • COCO2017_detection_91cl
    • coco_precision
    • 0.88%
    • 0.62%
    • 0.50%
    • faster_rcnn_resnet50_coco
    • COCO2017_detection_91cl_bkgr
    • coco_precision
    • 0.19%
    • 0.19%
    • 0.20%
    • googlenet-v4
    • ImageNet2012_bkgr
    • accuracy @ top1
    • 0.07%
    • 0.09%
    • 0.26%
    • mobilenet-ssd
    • VOC2007_detection
    • map
    • 0.47%
    • 0.14%
    • 0.48%
    • mobilenet-v2
    • ImageNet2012
    • accuracy @ top1
    • 0.50%
    • 0.18%
    • 0.20%
    • resnet-18
    • ImageNet2012
    • accuracy @ top1
    • 0.27%
    • 0.24%
    • 0.29%
    • resnet-50
    • ImageNet2012
    • accuracy @ top1
    • 0.13%
    • 0.12%
    • 0.13%
    • ssd-resnet34-1200
    • COCO2017_detection_80cl_bkgr
    • map
    • 0.08%
    • 0.09%
    • 0.06%
    • unet-camvid-onnx-0001
    • CamVid_12cl
    • mean_iou @ mean
    • 0.33%
    • 0.33%
    • 0.30%
    • yolo_v3_tiny
    • COCO2017_detection_80cl
    • map
    • 0.01%
    • 0.07%
    • 0.12%
    • yolo_v4
    • COCO2017_detection_80cl
    • map
    • 0.05%
    • 0.06%
    • 0.01%

@endsphinxdirective