Skip to content

Python script to generate prototxt on Caffe, inception_v3 \ inception_v4 \ inception_resnet \ fractalnet \ resnext

Notifications You must be signed in to change notification settings

xuexuetong1993/caffe-model

 
 

Repository files navigation

Caffe-model

Python script to generate prototxt on Caffe, specially the inception_v3\inception_v4\inception_resnet\fractalnet

Generator scripts

The prototxts can be visualized by ethereon.

Every model has a bn (batch normalization) version (maybe only bn version), the paper is Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Classificaiton (imagenet)

Introduction

This folder contains the deploy files(include generator scripts) and pre-train models of resnet-v1, resnet-v2, inception-v3, inception-resnet-v2 and densenet(coming soon).

We didn't train any model from scratch, some of them are converted from other deep learning framworks (inception-v3 from mxnet, inception-resnet-v2 from tensorflow), some of them are converted from other modified caffe (resnet-v2). But to achieve the original performance, finetuning is performed on imagenet for several epochs.

The main contribution belongs to the authors and model trainers.

Performance on imagenet

  1. Top-1/5 accuracy of pre-train models in this repository.
Network 224/299(single-crop) 224/299(12-crop) 320/395(single-crop) 320/395(12-crop)
resnet101-v2 78.05/93.88 80.01/94.96 79.63/94.84 80.71/95.43
resnet152-v2 79.15/94.58 80.76/95.32 80.34/95.26 81.16/95.68
resnet269-v2 80.29/95.00 81.75/95.80 81.30/95.67 82.13/96.15
inception-v3 78.33/94.25 80.40/95.27 79.90/95.18 80.75/95.76
inception-v4 79.97/94.91 81.40/95.70 81.32/95.68 81.88/96.08
inception-resnet-v2 80.14/95.17 81.54/95.92 81.25/95.98 81.85/96.29
resnext50_32x4d 77.63/93.69 79.47/94.65 78.90/94.47 79.63/94.97
resnext101_32x4d 78.70/94.21 80.53/95.11 80.09/95.03 80.81/95.41
resnext101_64x4d 79.40/94.59 81.12/95.41 80.74/95.37 81.52/95.69
wrn50_2(resnet50_1x128d) 77.87/93.87 79.91/94.94 79.32/94.72 80.17/95.13
  • The pre-train models are tested on original caffe by evaluation_cls.py, but ceil_mode:false(pooling_layer) is used for the models converted from torch, the detail in https://github.com/BVLC/caffe/pull/3057/files. If you remove ceil_mode:false, the performance will decline about 1% top1.
  • 224x224(base_size=256) and 320x320(base_size=320) crop size for resnet-v2/resnext/wrn, 299x299(base_size=320) and 395x395(base_size=395) crop size for inception.
  1. Top-1/5 accuracy with different crop sizes. teaser
  • Figure: Accuracy curves of inception_v3(left) and resnet101_v2(right) with different crop sizes.
  1. Download url and forward time cost for each model.

Forward time cost is evaluated with one image/mini-batch using cuDNN 5.1 on a Pascal Titan X GPU.

Network 224/299 320/395 Download(BaiDuCloud) Source
resnet101-v2 55.7ms 60.2ms caffemodel (170.3MB) craftGBD
resnet152-v2 82.8ms 89.1ms caffemodel (230.2MB) craftGBD
resnet269-v2 144.3ms 154.8ms caffemodel (390.4MB) craftGBD
inception-v3 56.2ms 60.4ms caffemodel (91.1MB) mxnet
inception-v4 91.9ms 96.2ms caffemodel (163.1MB) tensorflow_slim
inception-resnet-v2 127.1ms 133.6ms caffemodel (213.4MB) tensorflow_slim
resnext50_32x4d 37.0ms 41.2ms caffemodel (95.8MB) facebookresearch
resnext101_32x4d 69.4ms 74.4ms caffemodel (169.1MB) facebookresearch
resnext101_64x4d 79.1ms 87.6ms caffemodel (319.2MB) facebookresearch
wrn50_2(resnet50_1x128d) 33.4ms 37.1ms caffemodel (263.1MB) szagoruyko

Check the performance

  1. Download the ILSVRC 2012 classification val set 6.3GB, and put the extracted images into the directory:

    ~/Database/ILSVRC2012
    
  2. Check the resnet-v2 (101, 152 and 269) performance, the settings of evaluation_cls.py:

    val_file = 'ILSVRC2012_val.txt' # download from this folder, label range 0~999
    ... ...
    model_weights = 'resnet-v2/resnet101_v2.caffemodel' # download as below
    model_deploy = 'resnet-v2/deploy_resnet101_v2.prototxt' # check the parameters of input_shape
    ... ...
    mean_value = np.array([102.9801, 115.9465, 122.7717])  # BGR
    std = np.array([1.0, 1.0, 1.0])  # BGR
    crop_num = 1    # perform center(single)-crop
    

    Check the inception-v3 performance, the settings of evaluation_cls.py:

    val_file = 'ILSVRC2015_val.txt' # download from this folder, label range 0~999
    ... ...
    model_weights = 'inception_v3/inception_v3.caffemodel' # download as below
    model_deploy = 'inception_v3/deploy_inception_v3.prototxt' # check the parameters of input_shape
    ... ...
    mean_value = np.array([128.0, 128.0, 128.0])  # BGR
    std = np.array([128.0, 128.0, 128.0])  # BGR
    crop_num = 1    # perform center(single)-crop
    

    Check the inception-resnet-v2 (inception-v4) performance, the settings of evaluation_cls.py:

    val_file = 'ILSVRC2012_val.txt' # download from this folder, label range 0~999
    ... ...
    model_weights = 'inception_resnet_v2/inception_resnet_v2.caffemodel' # download as below
    model_deploy = 'inception_resnet_v2/deploy_inception_resnet_v2.prototxt' # check the parameters of input_shape
    ... ...
    mean_value = np.array([128.0, 128.0, 128.0])  # BGR
    std = np.array([128.0, 128.0, 128.0])  # BGR
    crop_num = 1    # perform center(single)-crop
    

    Check the resnext (50_32x4d, 101_32x4d and 101_64x4d) or wrn50_2 performance, the settings of evaluation_cls.py:

    val_file = 'ILSVRC2012_val.txt' # download from this folder, label range 0~999
    ... ...
    model_weights = 'resnext/resnext50/resnext50_32x4d.caffemodel' # download as below
    

   model_deploy = 'resnext/resnext50/deploy_resnext50_32x4d.prototxt' # check the parameters of input_shape ... ... mean_value = np.array([103.52, 116.28, 123.675]) # BGR std = np.array([57.375, 57.12, 58.395]) # BGR crop_num = 1 # perform center(single)-crop ```

  1. then
    python evaluation_cls.py
    

Acknowlegement

I greatly thank Yangqing Jia and BVLC group for developing Caffe

And I would like to thank all the authors of every cnn model

About

Python script to generate prototxt on Caffe, inception_v3 \ inception_v4 \ inception_resnet \ fractalnet \ resnext

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%