/pytorch-mobile

mobile vision

Primary LanguagePythonMIT LicenseMIT

目的：专用于集成各网络主干

clone from：https://github.com/bearpaw/pytorch-classification/blob/master/TRAINING.md （cifar&&imagenet&&curves）

Install

  • Install PyTorch
  • Clone recursively 
    git clone --recursive git@github.com:lcaikk1314/pytorch-classification.git

文件组织结构

models : 模型定义 models_init : 初始模型，从别的github获取的 models_onnx : 转出的ONNX模型存储位置

模型添加规则

按文件组织结构操作

处理流程

mobilev3_train.sh : 训练调优模型使用脚本

python3 imagenet.py
--arch MobileNetV3
--data ../imagenet/
--epochs 30 --schedule 11 21 --lr 0.0005 --gpu-id 0,1
--model_path './models_init/mbv3_large.pth.tar'
--checkpoint checkpoints_MobileNetV3_Large/imagenet/MobileNetV3

mobilev3_evaluate.sh : 模型评估，评估的指标：精度、计算量、参数量

python3 imagenet.py
--arch MobileNetV3
--data ../imagenet/
--gpu-id 0,1,2,3
--model_path './models_init/mbv3_large.pth.tar'
--evaluate

mobilev3_convert.sh : 模型转化为ONNX使用；

python3 convertonnx.py
--arch MobileNetV3
--model_path './models_init/mbv3_large.pth.tar'
--dest_path './models_onnx/mbv3_large.onnx'

models expert

Model Params (M) MFLOPS model size(M) Top-1 Error(%) Top-5 Error(%) 1threadtimes(ms) 2threadtimes(ms) VSS(K) RSS(K)
mobileV3-large 3.96 271.036 30.3 29.212 10.59 89.649 59.5683 61916 49112
mobileV3-small 2.51 65.558 19.2 35.074 14.534 29.9176 18.8802 42972 33668
efficientnet-b0 5.289 401.679 20.4 23.57 6.952 145.293 134.429 103900 82436
efficientnet-b1 7.794 591.948 30.1 21.628 5.964 236.81 202.56 122844 103528
efficientnet-b2 9.110 682.357 35.1 23.04 6.81 296.041 222.683 137692 114812
efficientnet-lite0 4.652 398.727 17.9 24.606 7.500 109.905 63.9615 69596 52736
regnet200m 4.151 255.463 15.9 29.91 10.466 63.1528 37.8787 69084 45972
regnet400m 5.74 513.125 22.0 26.21 8.282 141.344 76.282 83932 65632
regnet600m 6.633 595.577 25.4 25.27 7.648
ghostnetv1 5.183 148.788 19.9 26.02 8.538 63.1484 36.6433 72292 55764
Shufflenet_V2_1.0 2.28 148.809 8.78 30.64 11.684 45.672 29.876 37340 27404

models NAS

Model Params (M) MFLOPS model size(M) Top-1 Error (%) Top-5 Error (%)

pytorch-classification

Classification on CIFAR-10/100 and ImageNet with PyTorchV1.0.

Features

  • Unified interface for different network architectures
  • Multi-GPU support
  • Training progress bar with rich info
  • Training log and training curve visualization code (see ./utils/logger.py)

Install

  • Install PyTorch
  • Clone recursively 
    git clone --recursive https://github.com/bearpaw/pytorch-classification.git

Training

Please see the Training recipes for how to train the models.

Results

CIFAR

Top1 error rate on the CIFAR-10/100 benchmarks are reported. You may get different results when training your models with different random seed. Note that the number of parameters are computed on the CIFAR-10 dataset.

Model Params (M) CIFAR-10 (%) CIFAR-100 (%)
alexnet 2.47 22.78 56.13
vgg19_bn 20.04 6.66 28.05
ResNet-110 1.70 6.11 28.86
PreResNet-110 1.70 4.94 23.65
WRN-28-10 (drop 0.3) 36.48 3.79 18.14
ResNeXt-29, 8x64 34.43 3.69 17.38
ResNeXt-29, 16x64 68.16 3.53 17.30
DenseNet-BC (L=100, k=12) 0.77 4.54 22.88
DenseNet-BC (L=190, k=40) 25.62 3.32 17.17

cifar

ImageNet

Single-crop (224x224) validation error rate is reported.

Model Params (M) Top-1 Error (%) Top-5 Error (%)
ResNet-18 11.69 30.09 10.78
ResNeXt-50 (32x4d) 25.03 22.6 6.29

Validation curve

Pretrained models

Our trained models and training logs are downloadable at OneDrive.

Supported Architectures

CIFAR-10 / CIFAR-100

Since the size of images in CIFAR dataset is 32x32, popular network structures for ImageNet need some modifications to adapt this input size. The modified models is in the package models.cifar:

ImageNet

Contribute

Feel free to create a pull request if you find any bugs or you want to contribute (e.g., more datasets and more network structures).