ImageNet

Network Name Accuracy (top1) Quantized Input Resolution (HxWxC) Params (M) FLOPs (G) Pretrained Source Compiled
efficientnet_l 80.47 79.07 300x300x3 10.55 9.70 link link link
efficientnet_m🚀 78.98 78.35 240x240x3 6.87 3.68 link link link
efficientnet_s 77.61 76.93 224x224x3 5.41 2.36 link link link
efficientnet_lite0 74.93 74.25 224x224x3 4.63 0.39 link link link
efficientnet_lite1 76.66 76.16 240x240x3 5.39 0.61 link link link
efficientnet_lite2 77.44 76.23 260x260x3 6.06 0.87 link link link
efficientnet_lite3 79.2 78.49 280x280x3 8.16 1.40 link link link
efficientnet_lite4 80.79 80.01 300x300x3 12.95 2.58 link link link
hardnet39ds 73.44 71.81 224x224x3 3.48 0.43 link link link
hardnet68 75.48 75.06 224x224x3 17.56 4.25 link link link
inception_v1 69.76 69.55 224x224x3 6.62 1.50 link link link
mobilenet_v1 71.02 70.18 224x224x3 4.22 0.57 link link link
mobilenet_v2_1.0🚀 71.84 71.13 224x224x3 3.49 0.31 link link link
mobilenet_v2_1.4 74.11 73.15 224x224x3 6.09 0.59 link link link
mobilenet_v3 72.27 71.79 224x224x3 4.07 1.00 link link link
mobilenet_v3_large_minimalistic🚀 72.12 71.14 224x224x3 3.91 0.21 link link link
regnetx_1.6gf 77.07 76.47 224x224x3 9.17 1.61 link link link
regnetx_800mf🚀 75.07 74.42 224x224x3 7.24 0.80 link link link
regnety_200mf 70.32 69.85 224x224x3 3.15 0.20 link link link
resmlp12_relu 75.26 74.06 224x224x3 15.77 3.02 link link link
resnet_v1_18 71.21 70.61 224x224x3 11.68 1.82 link link link
resnet_v1_34 72.68 71.81 224x224x3 21.79 3.67 link link link
resnet_v1_50🚀⭐ 75.21 74.61 224x224x3 25.53 3.49 link link link
resnet_v2_18 69.58 68.2 224x224x3 11.68 1.82 link link link
resnet_v2_34 73.1 72.63 224x224x3 21.79 3.67 link link link
resnext26_32x4d 76.08 74.76 224x224x3 15.37 2.48 link link link
resnext50_32x4d 79.32 78.4 224x224x3 24.99 4.24 link link link
shufflenet_g8_w1 66.29 65.44 224x224x3 2.46 0.18 link link link
squeezenet_v1.1 59.88 59.02 224x224x3 1.24 0.39 link link link
  • Network available in Hailo Benchmark are marked with 🚀
  • Networks available in TAPPAS are marked with ⭐