Skip to content

mingxingtan/mnasnet

Repository files navigation

MNasNet

[1] Mingxing Tan, et al. MnasNet: Platform-Aware Neural Architecture Search for Mobile. CVPR 2019. Arxiv link: https://arxiv.org/pdf/1807.11626.pdf

About the Model

We provide a few standard-size and small-size AutoML models in mnasnet_models.py including:

  • mnasnet-a1 has ~75.2% top-1 ImageNet accuracy with 3.9M parameters and 312M Multiply-Adds.
  • mnasnet-small has ~66% top-1 ImageNet accuracy with 2.0M parameters and 68M Multiply-Adds.

The standard size MnasNet-A1 inference has 1.8x faster throughput (55% lower latency) than the corresponding MobileNetV2 model.

MnasNet-A1 and MobileNetV2

Comparing to MobileNetV2, MnasNet-A1 model has clear better performance in accuracy when they are at the same latency level.

MnasNet-A1 and MobileNetV2 Details

Here are the details of Mnasnet-A1 on ImageNet:

Input Size Depth Multiplier Top-1 Accuracy Top-5 Accuracy Parameters(M) Multi-Adds (M) Pixel 1 latency (ms)
224 1.4 77.2 93.5 6.1 591.5 135
224 1 75.2 92.5 3.9 315.2 78
224 0.75 73.3 91.3 2.9 226.7 61
224 0.5 68.9 88.4 2.1 105.2 32
224 0.35 64.1 85.1 1.7 63.2 22
192 1.4 76.1 93.0 6.1 435.1 99
192 1 74.0 91.6 3.9 232.0 57
192 0.75 72.1 90.5 2.9 166.9 45
192 0.5 67.2 87.4 2.1 77.6 24
192 0.35 62.4 83.8 1.7 46.8 17
160 1.4 74.8 92.1 6.1 302.8 72
160 1 72.0 90.5 3.9 161.6 41
160 0.75 70.1 89.3 2.9 116.4 33
160 0.5 64.9 85.8 2.1 54.4 18
160 0.35 52.3 81.5 1.7 32.9 13
128 1.4 72.5 90.6 6.1 194.5 49
128 1 69.3 88.9 3.9 104.1 29
128 0.75 67.0 87.3 2.9 75.0 23
128 0.5 60.8 83.0 2.1 35.3 12
128 0.35 54.8 78.1 1.7 21.6 8.5
96 1.4 68.6 88.1 6.1 110.3 32
96 1 64.4 85.8 3.9 59.3 18
96 0.75 62.1 84.0 2.9 42.9 17
96 0.5 54.7 78.1 2.1 20.5 7.4
96 0.35 49.3 73.4 1.7 12.7 5.4

For more information about training, please refer to our tutorial: https://cloud.google.com/tpu/docs/tutorials/mnasnet

About

MnasNet snapshot

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published