Skip to content

Latest commit

 

History

History
83 lines (67 loc) · 12.1 KB

checkpoints.md

File metadata and controls

83 lines (67 loc) · 12.1 KB
Raw

Benchmarks and Checkpoints

Each zip file contains 4 types of files

  • a checkpoint of the model, typically, named as model_best.pth.tar
  • the md5 of the checkpoint
  • a hyper-parameter json file, typically, named as hparams_train.json
  • tensorboard log file, you can use tensorboard to visualize the log. It is in the val directory within the zip file.

Hope you have fun with these checkpoints. Any issues about checkpoints should be raised at checkpoints.

ImageNet

General training protocols: batch size 256, epochs 120, cos learning rate 0.1, AutoAugment/RandAugment, Label smoothing, mixup, random erasing.

Methods Top-1/Top-5 Acc MParams/GFLOPs Checkpoints
ResNet-50, 224px 78.84 / 94.47 25.7 / 5.5 resnet50_split1_imagenet_256_06
SE-ResNet-50, 224px 79.47 / 94.54 28.2 / 4.9 se_resnet50_split1_imagenet_256_01
ResNeXSt-50, 4x16d, 224px 79.85 / 94.98 17.8 / 4.3 resnexst50_4x16d_split1_imagenet_256_01
ResNeXSt-50, 8x16d, 224px 80.90 / 95.36 30.5 / 6.8 resnexst50_8x16d_split1_imagenet_256_03
ResNeXSt-50, 4x32d, 224px 81.10 / 95.49 37.1 / 8.3 resnexst50_4x32d_split1_imagenet_256_05
ResNet-110, 224px 80.16 / 94.54 44.8 / 9.2 resnet101_split1_imagenet_256_01
WRN-50-2, 224px 80.66 / 95.16 68.9 / 12.8 wide_resnet50_2_split1_imagenet_256_01
WRN-50-2, S=2, 224px 79.64 / 94.82 51.4 / 10.9 wide_resnet50_2_split2_imagenet_256_02
WRN-50-3, 224px 80.74 / 95.40 135.0 / 23.8 wide_resnet50_3_split1_imagenet_256_01
WRN-50-3, S=2, 224px 81.42 / 95.62 138.0 / 25.6 wide_resnet50_3_split2_imagenet_256_02
ResNeXt-101, 64x4d, 224px 81.57 / 95.73 83.6 / 16.9 resnext101_64x4d_split1_imagenet_256_01
ResNeXt-101, 64x4d, S=2, 224px 82.13 / 95.98 88.6 / 18.8 resnext101_64x4d_split2_imagenet_256_02
EfficientNet-B7, 320px 81.83 / 95.78 66.7 / 10.6 efficientnetb7_split1_imagenet_128_03
EfficientNet-B7, S=2, 320px 82.74 / 96.30 68.2 / 10.5 efficientnetb7_split2_imagenet_128_02
SE-ResNeXt-101, 64x4d, S=2, 416px, 120 epochs 83.34 / 96.61 98.0 / 61.1 se_resnext101_64x4d_split2_imagenet_128_02
SE-ResNeXt-101, 64x4d, S=2, 320px, 350 epochs 83.60 / 96.69 98.0 / 38.2 se_resnext101_64x4d_B_split2_imagenet_128_05

CIFAR-100

Methods Top-1 Acc MParams/GFLOPs Checkpoints
WRN-28-10 84.50 36.5 / 5.25 wide_resnet28_10_split1_cifar100_128_01_acc84.5
WRN-28-10, S=2 85.52 35.8 / 5.16 wide_resnet28_10_split2_cifar100_128_02_acc85.52
WRN-28-10, S=4 85.74 36.7 / 5.28 wide_resnet28_10_split4_cifar100_128_03_acc85.74
WRN-40-10 83.98 55.9 / 8.08 wide_resnet40_10_split1_cifar100_128_06_acc83.98
WRN-40-10, S=2 85.91 54.8 / 7.94 wide_resnet40_10_split2_cifar100_128_05_acc85.91
WRN-40-10, S=4 86.90 56.0 / 8.12 wide_resnet40_10_split4_cifar100_128_04_acc86.90
DenseNet-BC-190 85.90 25.8 / 9.39 densenet190_split1_cifar100_64_01_acc85.90
DenseNet-BC-190, S=2 87.36 25.5 / 9.24 densenet190_split2_cifar100_128_02_acc87.36
DenseNet-BC-190, S=4 87.44 26.3 / 9.48 densenet190_split4_cifar100_64_03_acc87.44
PyramidNet-272 88.98 26.8 / 4.55 pyramidnet272_split1_cifar100_128_01_88.98
PyramidNet-272, S=2 89.25 28.9 / 5.24 pyramidnet272_split2_cifar100_128_06_acc89.25
PyramidNet-272, S=4 89.46 32.8 / 6.33 pyramidnet272_split4_cifar100_128_07_acc89.46

CIFAR-10

Methods Top-1 Acc MParams/GFLOPs Checkpoints
WRN-28-10 97.59 36.5 / 5.25 wide_resnet28_10_split1_cifar10_128_08_acc97.59
WRN-28-10, S=2 98.19 35.8 / 5.16 wide_resnet28_10_split2_cifar10_128_07_acc98.19
WRN-28-10, S=4 98.32 36.5 / 5.28 wide_resnet28_10_split4_cifar10_128_24_acc98.32
WRN-40-10 97.81 55.8 / 8.08 wide_resnet40_10_split1_cifar10_128_04_acc97.81
WRN-40-10, S=4 98.38 55.9 / 8.12 wide_resnet40_10_split4_cifar10_128_05_acc98.38
Shake-Shake 26 2x96d 98.00 26.2 / 3.78 shake_resnet26_2x96d_split1_cifar10_128_07_acc98.00
Shake-Shake 26 2x96d, S=2 98.25 23.3 / 3.38 shake_resnet26_2x96d_split2_cifar10_128_12
Shake-Shake 26 2x96d, S=4 98.31 26.3 / 3.81 shake_resnet26_2x96d_split4_cifar10_128_09
PyramidNet-272 98.67 26.2 / 4.55 pyramidnet272_split1_cifar10_128_01_acc98.67
PyramidNet-272, S=4 98.71 32.6 / 6.33 pyramidnet272_split4_cifar10_128_05_acc98.71

COCO val2017 (SSD300)

Backbones mAP MParams/GFLOPs Checkpoints
WRN-50-2 30.3 39.3 / 48.1 ssd300_coco_ssdv2_02
WRN-50-2, S=2 29.9 31.7 / 45.3 ssd300_coco_ssdv2_06
WRN-50-3 30.7 64.1 / 86.8 ssd300_coco_ssdv2_07
WRN-50-3, S=2 31.6 64.3 / 96.3 ssd300_coco_ssdv2_09
ResNeXt-101, 64x4d 32.6 68.9 / 90.1 ssd300_coco_ssdv2_14
ResNeXt-101, 64x4d, S=2 34.1 69.8 / 100.5 ssd300_coco_ssdv2_15