onnx model zoo准确率复现

本文对比了多种经典深度学习模型在ImageNet数据集上的官方及自测准确率,并记录了不同模型转换为ONNX格式后的版本信息及自测时采用的预处理方式。涉及模型包括AlexNet、VGG19、GoogleNet等。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

Modelfromversion官方准确率top5官方准确率top1自测准确率top5自测准确率top1diftop5diff top1自测使用的预处理方式
alexnetonnx model zooir3/opset880.20%57.10%78.15%54.50%2.05%2.60%bgr,mean:103.939 ,116.779,123.68
caffenetonnx model zooir3/opset880.40%57.40%79.25%55.66%1.15%1.74%bgr; mean 103.939 ,116.779,123.68
vgg19onnx model zooir3/opset891.58%73.72%88.86%69.09%2.72%4.63%resize*到333,crop 224; bgr; mean 103.939 ,116.779,123.68
googleNetonnx model zooir3/opset888.90%68.70%88.32%67.66%0.58%1.04%bgr; mean 103.939 ,116.779,123.68
inception_v1onnx model zooir3/opset889.60%69.80%87.81%67.16%1.79%2.64%bgr; mean 103.939 ,116.779,123.68
inception_v2onnx model zooir3/opset891.80%73.90%?
mobilenet_v2onnx model zooir3/opset789.99%70.94%88.50%68.35%1.49%2.59%rgb; mean 123.675,116.28,103.53; scale 58.395,57.12,57.375
shufflenet_v2onnx model zooir6/opset1088.32%69.56%86.68%66.43%1.64%3.13%rgb; reisze到256,crop224; mean 123.675,116.28,103.53; scale 58.395,57.12,57.375
resnet50_v2onnx model zooir3/opset792.38%74.93%91.86%73.79%0.52%1.14%rgb; mean 123.675,116.28,103.53; scale 58.395,57.12,57.375
resnet101_v2onnx model zooir3/opset793.20%76.48%92.07%74.44%1.13%2.04%rgb; mean 123.675,116.28,103.53; scale 58.395,57.12,57.375
desnet121onnx model zooir3/opset892.30%75.00%91.19%72.72%1.11%2.28%bgr; mean 103.53,116.28,123.675; scale 57.375,57.12,58.395
squeezenetonnx model zooir3/opset879.12%56.34%79.65%56.27%-0.53%0.07%bgr; mean 103.939 ,116.779,123.68
inception_v3https://github.com/Cadene/pretrained-models.pytorch#inceptionir4/opset993.45%77.29%93.31%76.69%0.14%0.61%input 1x3x299x299; rgb; mean 123.675,116.28,103.53; scale 58.395,57.12,57.375
inception_v4https://github.com/Cadene/pretrained-models.pytorch#inceptionir4/opset994.93%80.06%93.66%77.12%1.26%2.94%input 1x3x299x299; rgb; mean 123.675,116.28,103.53; scale 58.395,57.12,57.375
xceptionhttps://github.com/Cadene/pretrained-models.pytorch#inceptionir4/opset994.29%78.89%92.11%74.87%2.18%4.01%resize到333, center crop 224*224, x = x/255, x=(x-0.5)/0.5

转载请注明出处:https://blog.youkuaiyun.com/tbl1234567.作者:陶表犁

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值