Model Zoo

This page lists model archives that are pre-trained and pre-packaged, ready to be served for inference with MMS. To propose a model for inclusion, please submit a pull request.

Special thanks to the Apache MXNet community whose Model Zoo and Model Examples were used in generating these model archives.

Model File	Type	Dataset	Source	Size	Download
AlexNet	Image Classification	ImageNet	ONNX	233 MB	.mar
ArcFace-ResNet100	Face Recognition	Refined MS-Celeb1M	ONNX	236.4 MB	.mar
Character-level Convolutional Networks for Text Classification	Text Classification	Amazon Product Data	Gluon	40 MB	.mar
CaffeNet	Image Classification	ImageNet	MXNet	216 MB	.mar
FERPlus	Emotion Detection	FER2013	ONNX	35MB	.mar
Inception v1	Image Classification	ImageNet	ONNX	27 MB	.mar
Inception v3 w/BatchNorm	Image Classification	ImageNet	MXNet	45 MB	.mar
LSTM PTB	Language Modeling	PennTreeBank	MXNet	16 MB	.mar
MobileNetv2-1.0	Image Classification	ImageNet	ONNX	13.7 MB	.mar
Network in Network (NiN)	Image Classification	ImageNet	MXNet	30 MB	.mar
ResNet-152	Image Classification	ImageNet	MXNet	241 MB	.mar
ResNet-18	Image Classification	ImageNet	MXNet	43 MB	.mar
ResNet50-SSD	SSD (Single Shot MultiBox Detector)	ImageNet	MXNet	124 MB	.mar
ResNext101-64x4d	Image Classification	ImageNet	MXNet	334 MB	.mar
ResNet-18v1	Image Classification	ImageNet	ONNX	45 MB	.mar
ResNet-34v1	Image Classification	ImageNet	ONNX	83 MB	.mar
ResNet-50v1	Image Classification	ImageNet	ONNX	98 MB	.mar
ResNet-101v1	Image Classification	ImageNet	ONNX	171 MB	.mar
ResNet-152v1	Image Classification	ImageNet	ONNX	231 MB	.mar
ResNet-18v2	Image Classification	ImageNet	ONNX	45 MB	.mar
ResNet-34v2	Image Classification	ImageNet	ONNX	83 MB	.mar
ResNet-50v2	Image Classification	ImageNet	ONNX	98 MB	.mar
ResNet-101v2	Image Classification	ImageNet	ONNX	171 MB	.mar
ResNet-152v2	Image Classification	ImageNet	ONNX	231 MB	.mar
Shufflenet	Image Classification	ImageNet	ONNX	8.1 MB	.mar
SqueezeNet_v1.1	Image Classification	ImageNet	ONNX	5 MB	.mar
SqueezeNet v1.1	Image Classification	ImageNet	MXNet	5 MB	.mar
VGG16	Image Classification	ImageNet	MXNet	490 MB	.mar
VGG16	Image Classification	ImageNet	ONNX	527 MB	.mar
VGG16_bn	Image Classification	ImageNet	ONNX	527 MB	.mar
VGG19	Image Classification	ImageNet	MXNet	509 MB	.mar
VGG19	Image Classification	ImageNet	ONNX	548 MB	.mar
VGG19_bn	Image Classification	ImageNet	ONNX	548 MB	.mar

Details on Each Model

Each model below comes with a basic description, and where available, a link to a scholarly article about the model.

Many of these models use a kitten image to test inference. Use the following to get one that will work:

curl -O https://s3.amazonaws.com/model-server/inputs/kitten.jpg

AlexNet

• Type: Image classification trained on ImageNet

• Reference: Krizhevsky, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models alexnet=https://s3.amazonaws.com/model-server/model_archive_1.0/alexnet.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/alexnet -T kitten.jpg
  

ArcFace-ResNet100 (from ONNX model zoo)

• Type: Face Recognition model trained on refined MS-Celeb1M dataset (model imported from ONNX)

• Reference: Deng et al.

• Model Service:
  • arcface_service.py
  • mtcnn_detector.py
• Install dependencies:

  pip install opencv-python
  pip install scikit-learn
  pip install easydict
  pip install scikit-image
  pip install numpy
  

• Start Server:

  multi-model-server --start --models arcface=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-arcface-resnet100.mar
  

• Get two test images: ```bash curl -O https://s3.amazonaws.com/model-server/inputs/arcface-input1.jpg

curl -O https://s3.amazonaws.com/model-server/inputs/arcface-input2.jpg

* **Run Prediction**:
```bash
curl -X POST http://127.0.0.1:8080/predictions/arcface -F "img1=@arcface-input1.jpg" -F "img2=@arcface-input2.jpg"

CaffeNet

• Type: Image classification trained on ImageNet

• Reference: Krizhevsky, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models caffenet=https://s3.amazonaws.com/model-server/model_archive_1.0/caffenet.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/caffenet -T kitten.jpg
  

Character-level Convolutional Networks for text Classification

• Type: Character-level Convolutional network for text classification trained on Amazon Product Data.

• Reference: R. He, J. McAuley et al., J. McAuley, C. Targett, J. Shi, A. van den Hengel et al.

• Model Service: gluon_crepe.py

• Start Server:

  multi-model-server --start --models crepe=https://s3.amazonaws.com/model-server/model_archive_1.0/crepe.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/crepe -F "data=[{\"review_title\":\"Inception is the best\",\"review\": \"great direction and story\"}]"
  

DUC-ResNet101 (from ONNX model zoo)

• Type: Semantic Segmentation model trained on the Cityscapes dataset (model imported from ONNX)

• Reference: Wang et al.

• Model Service:
  • duc_service.py
  • cityscapes_labels.py
• Install dependencies:

  pip install opencv-python
  pip install pillow
  

• Start Server:

  multi-model-server --models duc=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-duc.mar
  

• Get the test image:

  curl -O https://s3.amazonaws.com/multi-model-server/onnx-duc/city1.jpg
  

• Download inference script:

The script makes an inference call to the server using the test image, displays the colorized segmentation map and prints the confidence score.

curl -O https://s3.amazonaws.com/multi-model-server/onnx-duc/duc-inference.py

• Run Prediction:

  python duc-inference.py city1.jpg
  

FERPlus

• Type: Emotion detection trained on FER2013 dataset (model imported from ONNX)

• Reference: Barsoum et al.

• Model Service: emotion_detection_service.py

• Start Server:

  multi-model-server --start --models FERPlus=https://s3.amazonaws.com/model-server/model_archive_1.0/FERPlus.mar
  

• Get a test image:

  curl -O https://s3.amazonaws.com/model-server/inputs/ferplus-input.jpg
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/FERPlus -T ferplus-input.jpg
  

Inception v1

• Type: Image classification trained on ImageNet

• Reference: Szegedy, et al., Google

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models onnx-inception-v1=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-inception_v1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/onnx-inception-v1 -T kitten.jpg
  

Inception v3

• Type: Image classification trained on ImageNet

• Reference: Szegedy, et al., Google

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models inception-bn=https://s3.amazonaws.com/model-server/model_archive_1.0/inception-bn.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/inception-bn -T kitten.jpg
  

LSTM PTB

Long short-term memory network trained on the PennTreeBank dataset.

• Reference: Hochreiter, et al.

• Model Service: lstm_ptb_service.py

• Start Server:

  multi-model-server --start --models lstm_ptb=https://s3.amazonaws.com/model-server/model_archive_1.0/lstm_ptb.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/lstm_ptb -H "Content-Type: application/json" -d '[{"input_sentence": "on the exchange floor as soon as ual stopped trading we <unk> for a panic said one top floor trader"}]'
  

MobileNetv2-1.0 (from ONNX model zoo)

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: Sandler et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models mobilenet=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-mobilenet.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/mobilenet -T kitten.jpg
  

Network in Network

• Type: Image classification trained on ImageNet

• Reference: Lin, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models nin=https://s3.amazonaws.com/model-server/model_archive_1.0/nin.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/nin -T kitten.jpg
  

ResNet-152

• Type: Image classification trained on ImageNet

• Reference: Lin, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet-152=https://s3.amazonaws.com/model-server/model_archive_1.0/resnet-152.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet-152 -T kitten.jpg
  

ResNet-18

• Type: Image classification trained on ImageNet

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet-18=https://s3.amazonaws.com/model-server/model_archive_1.0/resnet-18.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet-18 -T kitten.jpg
  

ResNet50-SSD

• Type: Image classification trained on ImageNet

• Reference: Liu, et al.

• Model Service: ssd_service.py

• Start Server:

  multi-model-server --start --models SSD=https://s3.amazonaws.com/model-server/model_archive_1.0/resnet50_ssd.mar
  

• Run Prediction: ```bash curl -O https://www.dphotographer.co.uk/users/21963/thm1024/1337890426_Img_8133.jpg

curl -X POST http://127.0.0.1:8080/predictions/SSD -T 1337890426_Img_8133.jpg

## <a name="resnext101"></a>ResNext101-64x4d
* **Type**: Image classification trained on ImageNet

* **Reference**: [Xie, et al.](https://arxiv.org/pdf/1611.05431.pdf)

* **Model Service**: [mxnet_vision_service.py](https://github.com/awslabs/multi-model-server/blob/master/examples/mxnet_vision/mxnet_vision_service.py)

* **Start Server**:
```bash
multi-model-server --start --models resnext101=https://s3.amazonaws.com/model-server/model_archive_1.0/resnext-101-64x4d.mar

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnext101 -T kitten.jpg
  

ResNet (from ONNX model zoo)

ResNet18-v1

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet18-v1=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet18v1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet18-v1 -T kitten.jpg
  

ResNet34-v1

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet34-v1=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet34v1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet34-v1 -T kitten.jpg
  

ResNet50-v1

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet50-v1=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet50v1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet50-v1 -T kitten.jpg
  

ResNet101-v1

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet101-v1=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet101v1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet101-v1 -T kitten.jpg
  

ResNet152-v1

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet152-v1=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet152v1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet152-v1 -T kitten.jpg
  

ResNet18-v2

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet18-v2=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet18v2.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet18-v2 -T kitten.jpg
  

ResNet34-v2

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet34-v2=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet34v2.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet34-v2 -T kitten.jpg
  

ResNet50-v2

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet50-v2=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet50v2.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet50-v2 -T kitten.jpg
  

ResNet101-v2

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet101-v2=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet101v2.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet101-v2 -T kitten.jpg
  

ResNet152-v2

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: He, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models resnet152-v2=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-resnet152v2.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/resnet152-v2 -T kitten.jpg
  

Shufflenet_v2

• Type: Image classification trained on ImageNet

• Reference: Zhang, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models shufflenet=https://s3.amazonaws.com/model-server/model_archive_1.0/shufflenet.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/shufflenet -T kitten.jpg
  

SqueezeNet v1.1

• Type: Image classification trained on ImageNet

• Reference: Iandola, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models squeezenet_v1.1=https://s3.amazonaws.com/model-server/model_archive_1.0/squeezenet_v1.1.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/squeezenet_v1.1 -T kitten.jpg
  

SqueezeNet v1.1 (from ONNX model zoo)

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: Iandola, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models onnx-squeezenet=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-squeezenet.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/onnx-squeezenet -T kitten.jpg
  

VGG16

• Type: Image classification trained on ImageNet

• Reference: Simonyan, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models vgg16=https://s3.amazonaws.com/model-server/model_archive_1.0/vgg16.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/vgg16 -T kitten.jpg
  

VGG19

• Type: Image classification trained on ImageNet

• Reference: Simonyan, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models vgg19=https://s3.amazonaws.com/model-server/model_archive_1.0/vgg19.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/vgg19 -T kitten.jpg
  

VGG (from ONNX model zoo)

VGG16

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: Simonyan, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models onnx-vgg16=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-vgg16.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/onnx-vgg16 -T kitten.jpg
  

VGG16_bn

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: Simonyan, et al. (Batch normalization applied after each conv layer of VGG16)

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models onnx-vgg16_bn=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-vgg16_bn.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/onnx-vgg16_bn -T kitten.jpg
  

VGG19

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: Simonyan, et al.

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models onnx-vgg19=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-vgg19.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/onnx-vgg19 -T kitten.jpg
  

VGG19_bn

• Type: Image classification trained on ImageNet (imported from ONNX)

• Reference: Simonyan, et al. (Batch normalization applied after each conv layer of VGG19)

• Model Service: mxnet_vision_service.py

• Start Server:

  multi-model-server --start --models onnx-vgg19_bn=https://s3.amazonaws.com/model-server/model_archive_1.0/onnx-vgg19_bn.mar
  

• Run Prediction:

  curl -X POST http://127.0.0.1:8080/predictions/onnx-vgg19_bn -T kitten.jpg