MMdnn/mmdnn/models

Introduction

This is a collection of pre-trained models in different deep learning frameworks.

You can download the model you want by simply click the download link.

With the download model, you can convert them to different frameworks.

Next session show an example to show you how to convert pre-trained model between frameworks.

Steps to Convert Model

Example: Convert vgg19 model from Tensorflow to CNTK

1. Install the stable version of MMdnn

   pip install mmdnn
   

2. Download Tensorflow pre-trained model

   • Method 1: Directly download from below model collection
   • Method 2: Use command line

       $ mmdownload -f tensorflow -n vgg19
   
       Downloading file [./vgg_19_2016_08_28.tar.gz] from [http://download.tensorflow.org/models/vgg_19_2016_08_28.tar.gz]
       progress: 520592.0 KB downloaded, 100%
       Model saved in file: ./imagenet_vgg19.ckpt
   

   NOTICE: the model name after the '-n' argument must be the models appearence in the below model collection.

3. Convert model architecture(*.ckpt.meta) and weights(.ckpt) from Tensorflow to IR

   $ mmtoir -f tensorflow -d vgg19 -n imagenet_vgg19.ckpt.meta -w imagenet_vgg19.ckpt  --dstNodeName MMdnn_Output
   
   Parse file [imagenet_vgg19.ckpt.meta] with binary format successfully.
   Tensorflow model file [imagenet_vgg19.ckpt.meta] loaded successfully.
   Tensorflow checkpoint file [imagenet_vgg19.ckpt] loaded successfully. [38] variables loaded.
   IR network structure is saved as [vgg19.json].
   IR network structure is saved as [vgg19.pb].
   IR weights are saved as [vgg19.npy].
   

4. Convert models from IR to PyTorch code snippet and weights

   $ mmtocode -f pytorch -n vgg19.pb --IRWeightPath vgg19.npy --dstModelPath pytorch_vgg19.py -dw pytorch_vgg19.npy
   
   Parse file [vgg19.pb] with binary format successfully.
   Target network code snippet is saved as [pytorch_vgg19.py].
   Target weights are saved as [pytorch_vgg19.npy].
   

5. Generate PyTorch model from code snippet file and weight file

   $ mmtomodel -f pytorch -in pytorch_vgg19.py -iw pytorch_vgg19.npy --o pytorch_vgg19.pth
   
   PyTorch model file is saved as [pytorch_vgg19.pth], generated by [pytorch_vgg19.py] and [pytorch_vgg19.npy].
   Notice that you may need [pytorch_vgg19.py] to load the model back.
   

Model Collection

Image Classification

imagenet

alexnet Framework: caffe Download: prototxt caffemodel Source: Link	inception_v1 Framework: caffe Download: prototxt caffemodel Source: Link	vgg16 Framework: caffe Download: prototxt caffemodel Source: Link
vgg19 Framework: caffe Download: prototxt caffemodel Source: Link	resnet50 Framework: caffe Download: prototxt caffemodel Source: Link	resnet101 Framework: caffe Download: prototxt caffemodel Source: Link
resnet152 Framework: caffe Download: prototxt caffemodel Source: Link	squeezenet Framework: caffe Download: prototxt caffemodel Source: Link	xception Framework: caffe Download: prototxt caffemodel Source:
inception_v4 Framework: caffe Download: prototxt caffemodel Source:	alexnet Framework: cntk Download: model Source: Link	inception_v3 Framework: cntk Download: model Source: Link
resnet18 Framework: cntk Download: model Source: Link	resnet50 Framework: cntk Download: model Source: Link	resnet101 Framework: cntk Download: model Source: Link
resnet152 Framework: cntk Download: model Source: Link	inception_v3 Framework: coreml Download: mlmodel Source:	vgg16 Framework: coreml Download: mlmodel Source: Link
resnet50 Framework: coreml Download: mlmodel Source: Link	mobilenet Framework: coreml Download: mlmodel Source: Link	imagenet1k-inception-bn Framework: mxnet Download: json params Source: Link
imagenet1k-resnet-18 Framework: mxnet Download: json params Source: Link	imagenet1k-resnet-34 Framework: mxnet Download: json params Source: Link	imagenet1k-resnet-50 Framework: mxnet Download: json params Source: Link
imagenet1k-resnet-101 Framework: mxnet Download: json params Source: Link	imagenet1k-resnet-152 Framework: mxnet Download: json params Source: Link	imagenet1k-resnext-50 Framework: mxnet Download: json params Source: Link
imagenet1k-resnext-101 Framework: mxnet Download: json params Source: Link	imagenet1k-resnext-101-64x4d Framework: mxnet Download: json params Source: Link	vgg19 Framework: mxnet Download: json params Source: Link
vgg16 Framework: mxnet Download: json params Source: Link	squeezenet_v1.0 Framework: mxnet Download: json params Source: Link	squeezenet_v1.1 Framework: mxnet Download: json params Source: Link
alexnet Framework: pytorch Download: pth Source: Link	densenet121 Framework: pytorch Download: pth Source: Link	densenet169 Framework: pytorch Download: pth Source: Link
densenet201 Framework: pytorch Download: pth Source: Link	densenet161 Framework: pytorch Download: pth Source: Link	inception_v3 Framework: pytorch Download: pth Source: Link
resnet18 Framework: pytorch Download: pth Source: Link	resnet34 Framework: pytorch Download: pth Source: Link	resnet50 Framework: pytorch Download: pth Source: Link
resnet101 Framework: pytorch Download: pth Source: Link	resnet152 Framework: pytorch Download: pth Source: Link	squeezenet1_0 Framework: pytorch Download: pth Source: Link
squeezenet1_1 Framework: pytorch Download: pth Source: Link	vgg11 Framework: pytorch Download: pth Source: Link	vgg13 Framework: pytorch Download: pth Source: Link
vgg16 Framework: pytorch Download: pth Source: Link	vgg19 Framework: pytorch Download: pth Source: Link	vgg11_bn Framework: pytorch Download: pth Source: Link
vgg13_bn Framework: pytorch Download: pth Source: Link	vgg16_bn Framework: pytorch Download: pth Source: Link	vgg19_bn Framework: pytorch Download: pth Source: Link
vgg16 Framework: tensorflow Download: tgz Source: Link	vgg19 Framework: tensorflow Download: tgz Source: Link	inception_v1 Framework: tensorflow Download: tgz Source: Link
inception_v1_frozen Framework: tensorflow Download: tgz Source: Link	inception_v3 Framework: tensorflow Download: tgz Source: Link	inception_v3_frozen Framework: tensorflow Download: tgz Source: Link
resnet_v1_50 Framework: tensorflow Download: tgz Source: Link	resnet_v1_152 Framework: tensorflow Download: tgz Source: Link	resnet_v2_50 Framework: tensorflow Download: tgz Source: Link
resnet_v2_152 Framework: tensorflow Download: tgz Source: Link	resnet_v2_200 Framework: tensorflow Download: tgz Source: Link	mobilenet_v1_1.0 Framework: tensorflow Download: tgz Source: Link
mobilenet_v1_1.0_frozen Framework: tensorflow Download: tgz Source: Link	mobilenet_v2_1.0_224 Framework: tensorflow Download: tgz Source: Link	inception_resnet_v2 Framework: tensorflow Download: tgz Source: Link
nasnet-a_large Framework: tensorflow Download: tgz Source: Link

imagenet11k

imagenet11k-resnet-152 Framework: mxnet Download: json params Source: Link	imagenet11k-place365ch-resnet-152 Framework: mxnet Download: json params Source: Link	imagenet11k-place365ch-resnet-50 Framework: mxnet Download: json params Source: Link

Object Detection

Pascal VOC

voc-fcn8s Framework: caffe Download: prototxt caffemodel Source: Link	voc-fcn16s Framework: caffe Download: prototxt caffemodel Source: Link	voc-fcn32s Framework: caffe Download: prototxt caffemodel Source: Link
Fast-RCNN_Pascal Framework: cntk Download: model Source: Link	tinyyolo Framework: coreml Download: mlmodel Source: Link	yolov3 Framework: darknet Download: cfg weights Source: Link
yolov2 Framework: darknet Download: cfg weights Source: Link

grocery100

Fast-RCNN_grocery100 Framework: cntk Download: model Source: Link