From 704d0d556f915c89d9c2e6e890d19739df9b7003 Mon Sep 17 00:00:00 2001
From: Xinyu Chen <xinyu1.chen@intel.com>
Date: Wed, 1 May 2019 05:59:34 +0800
Subject: [PATCH] [int8] Add MobileNetV2_1.0 & ResNet18 Quantization (#14823)

* add resnet18 and mobilenetv2 models

* add readme

* support mkldnn s8s8 goihw16g weight format

* fix_readme_typo
---
 example/quantization/README.md                | 49 ++++++++++---------
 .../quantization/imagenet_gen_qsym_mkldnn.py  | 20 ++++++--
 src/operator/nn/mkldnn/mkldnn_base.cc         |  1 +
 3 files changed, 43 insertions(+), 27 deletions(-)

diff --git a/example/quantization/README.md b/example/quantization/README.md
index fc9a26755b4e..c39d027f8aed 100644
--- a/example/quantization/README.md
+++ b/example/quantization/README.md
@@ -27,17 +27,19 @@ The following models have been tested on Linux systems.
 
 | Model | Source | Dataset | FP32 Accuracy (top-1/top-5)| INT8 Accuracy (top-1/top-5)|
 |:---|:---|---|:---:|:---:|
+| [ResNet18-V1](#3)  | [Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)  | [Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)  |70.07%/89.30%|69.85%/89.23%|
 | [ResNet50-V1](#3)  | [Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)  | [Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)  | 75.87%/92.72%  |  75.71%/92.65% |
-| [ResNet101-V1](#4)  | [Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)  | [Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)  | 77.3%/93.58%  | 77.09%/93.41%  |
-|[Squeezenet 1.0](#5)|[Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|57.01%/79.71%|56.62%/79.55%|
-|[MobileNet 1.0](#6)|[Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|69.76%/89.32%|69.61%/89.09%|
+| [ResNet101-V1](#3)  | [Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)  | [Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)  | 77.3%/93.58%  | 77.09%/93.41%  |
+|[Squeezenet 1.0](#4)|[Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|57.01%/79.71%|56.62%/79.55%|
+|[MobileNet 1.0](#5)|[Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|69.76%/89.32%|69.61%/89.09%|
+|[MobileNetV2 1.0](#6)|[Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|70.14%/89.60%|69.53%/89.24%|
 |[Inception V3](#7)|[Gluon-CV](https://gluon-cv.mxnet.io/model_zoo/classification.html)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|76.49%/93.10% |76.38%/93% |
 |[ResNet152-V2](#8)|[MXNet ModelZoo](http://data.mxnet.io/models/imagenet/resnet/152-layers/)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|76.76%/93.03%|76.48%/92.96%|
 |[Inception-BN](#9)|[MXNet ModelZoo](http://data.mxnet.io/models/imagenet/inception-bn/)|[Validation Dataset](http://data.mxnet.io/data/val_256_q90.rec)|72.09%/90.60%|72.00%/90.53%|
 | [SSD-VGG16](#10) | [example/ssd](https://github.com/apache/incubator-mxnet/tree/master/example/ssd)  | VOC2007/2012  | 0.8366 mAP  | 0.8364 mAP  |
 | [SSD-VGG16](#10) | [example/ssd](https://github.com/apache/incubator-mxnet/tree/master/example/ssd)  | COCO2014  | 0.2552 mAP  | 0.253 mAP  |
 
-<h3 id='3'>ResNet50-V1</h3>
+<h3 id='3'>ResNet18/50/101-V1</h3>
 
 The following command is to download the pre-trained model from Gluon-CV and transfer it into the symbolic model which would be finally quantized. The [validation dataset](http://data.mxnet.io/data/val_256_q90.rec) is available for testing the pre-trained models:
 
@@ -62,37 +64,36 @@ python imagenet_inference.py --symbol-file=./model/resnet50_v1-symbol.json --bat
 python imagenet_inference.py --symbol-file=./model/resnet50_v1-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
 ```
 
-<h3 id='4'>ResNet101-V1</h3>
+<h3 id='4'>SqueezeNet 1.0</h3>
 
 The following command is to download the pre-trained model from Gluon-CV and transfer it into the symbolic model which would be finally quantized. The [validation dataset](http://data.mxnet.io/data/val_256_q90.rec) is available for testing the pre-trained models:
 
 ```
-python imagenet_gen_qsym_mkldnn.py --model=resnet101_v1 --num-calib-batches=5 --calib-mode=naive
+python imagenet_gen_qsym_mkldnn.py --model=squeezenet1.0 --num-calib-batches=5 --calib-mode=naive
 ```
-
 The model would be automatically replaced in fusion and quantization format. It is then saved as the quantized symbol and parameter files in the `./model` directory. The following command is to launch inference.
 
 ```
 # USE MKLDNN AS SUBGRAPH BACKEND
 export MXNET_SUBGRAPH_BACKEND=MKLDNN
 
-# Launch FP32 Inference 
-python imagenet_inference.py --symbol-file=./model/resnet101_v1-symbol.json --param-file=./model/resnet101_v1-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu --data-nthreads=1
+# Launch FP32 Inference
+python imagenet_inference.py --symbol-file=./model/squeezenet1.0-symbol.json --param-file=./model/squeezenet1.0-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu --data-nthreads=1
 
 # Launch INT8 Inference
-python imagenet_inference.py --symbol-file=./model/resnet101_v1-quantized-5batches-naive-symbol.json --param-file=./model/resnet101_v1-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu  --data-nthreads=1
+python imagenet_inference.py --symbol-file=./model/squeezenet1.0-quantized-5batches-naive-symbol.json --param-file=./model/squeezenet1.0-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu  --data-nthreads=1
 
 # Launch dummy data Inference
-python imagenet_inference.py --symbol-file=./model/resnet101_v1-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
-python imagenet_inference.py --symbol-file=./model/resnet101_v1-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
+python imagenet_inference.py --symbol-file=./model/squeezenet1.0-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu  --benchmark=True
+python imagenet_inference.py --symbol-file=./model/squeezenet1.0-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
 ```
 
-<h3 id='5'>SqueezeNet 1.0</h3>
+<h3 id='5'>MobileNet 1.0</h3>
 
 The following command is to download the pre-trained model from Gluon-CV and transfer it into the symbolic model which would be finally quantized. The [validation dataset](http://data.mxnet.io/data/val_256_q90.rec) is available for testing the pre-trained models:
 
 ```
-python imagenet_gen_qsym_mkldnn.py --model=squeezenet1.0 --num-calib-batches=5 --calib-mode=naive
+python imagenet_gen_qsym_mkldnn.py --model=mobilenet1.0 --num-calib-batches=5 --calib-mode=naive
 ```
 The model would be automatically replaced in fusion and quantization format. It is then saved as the quantized symbol and parameter files in the `./model` directory. The following command is to launch inference.
 
@@ -101,22 +102,22 @@ The model would be automatically replaced in fusion and quantization format. It
 export MXNET_SUBGRAPH_BACKEND=MKLDNN
 
 # Launch FP32 Inference
-python imagenet_inference.py --symbol-file=./model/squeezenet1.0-symbol.json --param-file=./model/squeezenet1.0-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu --data-nthreads=1
+python imagenet_inference.py --symbol-file=./model/mobilenet1.0-symbol.json --param-file=./model/mobilenet1.0-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu --data-nthreads=1
 
 # Launch INT8 Inference
-python imagenet_inference.py --symbol-file=./model/squeezenet1.0-quantized-5batches-naive-symbol.json --param-file=./model/squeezenet1.0-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu  --data-nthreads=1
+python imagenet_inference.py --symbol-file=./model/mobilenet1.0-quantized-5batches-naive-symbol.json --param-file=./model/mobilenet1.0-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu  --data-nthreads=1
 
 # Launch dummy data Inference
-python imagenet_inference.py --symbol-file=./model/squeezenet1.0-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu  --benchmark=True
-python imagenet_inference.py --symbol-file=./model/squeezenet1.0-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
+python imagenet_inference.py --symbol-file=./model/mobilenet1.0-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu  --benchmark=True
+python imagenet_inference.py --symbol-file=./model/mobilenet1.0-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
 ```
 
-<h3 id='6'>MobileNet 1.0</h3>
+<h3 id='6'>MobileNetV2 1.0</h3>
 
 The following command is to download the pre-trained model from Gluon-CV and transfer it into the symbolic model which would be finally quantized. The [validation dataset](http://data.mxnet.io/data/val_256_q90.rec) is available for testing the pre-trained models:
 
 ```
-python imagenet_gen_qsym_mkldnn.py --model=mobilenet1.0 --num-calib-batches=5 --calib-mode=naive
+python imagenet_gen_qsym_mkldnn.py --model=mobilenetv2_1.0 --num-calib-batches=5 --calib-mode=naive
 ```
 The model would be automatically replaced in fusion and quantization format. It is then saved as the quantized symbol and parameter files in the `./model` directory. The following command is to launch inference.
 
@@ -125,14 +126,14 @@ The model would be automatically replaced in fusion and quantization format. It
 export MXNET_SUBGRAPH_BACKEND=MKLDNN
 
 # Launch FP32 Inference
-python imagenet_inference.py --symbol-file=./model/mobilenet1.0-symbol.json --param-file=./model/mobilenet1.0-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu --data-nthreads=1
+python imagenet_inference.py --symbol-file=./model/mobilenetv2_1.0-symbol.json --param-file=./model/mobilenetv2_1.0-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu --data-nthreads=1
 
 # Launch INT8 Inference
-python imagenet_inference.py --symbol-file=./model/mobilenet1.0-quantized-5batches-naive-symbol.json --param-file=./model/mobilenet1.0-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu  --data-nthreads=1
+python imagenet_inference.py --symbol-file=./model/mobilenetv2_1.0-quantized-5batches-naive-symbol.json --param-file=./model/mobilenetv2_1.0-quantized-0000.params --rgb-mean=123.68,116.779,103.939 --rgb-std=58.393,57.12,57.375 --num-skipped-batches=50 --batch-size=64 --num-inference-batches=500 --dataset=./data/val_256_q90.rec --ctx=cpu  --data-nthreads=1
 
 # Launch dummy data Inference
-python imagenet_inference.py --symbol-file=./model/mobilenet1.0-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu  --benchmark=True
-python imagenet_inference.py --symbol-file=./model/mobilenet1.0-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
+python imagenet_inference.py --symbol-file=./model/mobilenetv2_1.0-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu  --benchmark=True
+python imagenet_inference.py --symbol-file=./model/mobilenetv2_1.0-quantized-5batches-naive-symbol.json --batch-size=64 --num-inference-batches=500 --ctx=cpu --benchmark=True
 ```
 
 <h3 id='7'>Inception-V3</h3>
diff --git a/example/quantization/imagenet_gen_qsym_mkldnn.py b/example/quantization/imagenet_gen_qsym_mkldnn.py
index 2ef137273cca..06a1272caf21 100644
--- a/example/quantization/imagenet_gen_qsym_mkldnn.py
+++ b/example/quantization/imagenet_gen_qsym_mkldnn.py
@@ -92,11 +92,13 @@ def save_params(fname, arg_params, aux_params, logger=None):
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Generate a calibrated quantized model from a FP32 model with Intel MKL-DNN support')
-    parser.add_argument('--model', type=str, choices=['resnet50_v1',
+    parser.add_argument('--model', type=str, choices=['resnet18_v1',
+                                                      'resnet50_v1',
                                                       'resnet101_v1',
                                                       'inceptionv3',
                                                       'squeezenet1.0',
                                                       'mobilenet1.0',
+                                                      'mobilenetv2_1.0',
                                                       'imagenet1k-resnet-152',
                                                       'imagenet1k-inception-bn',
                                                       'custom'],
@@ -163,7 +165,13 @@ def save_params(fname, arg_params, aux_params, logger=None):
         download_calib_dataset('http://data.mxnet.io/data/val_256_q90.rec', args.calib_dataset)
 
     # download model
-    if args.model in ['resnet50_v1', 'resnet101_v1', 'squeezenet1.0', 'mobilenet1.0', 'inceptionv3']:
+    if args.model in ['resnet18_v1',
+                      'resnet50_v1',
+                      'resnet101_v1',
+                      'squeezenet1.0',
+                      'mobilenet1.0',
+                      'mobilenetv2_1.0',
+                      'inceptionv3']:
         logger.info('model %s is converted from GluonCV' % args.model)
         args.use_gluon_model = True
     if args.use_gluon_model == True:
@@ -216,7 +224,7 @@ def save_params(fname, arg_params, aux_params, logger=None):
         excluded_sym_names += ['flatten']
         if exclude_first_conv:
             excluded_sym_names += ['conv_1']
-    elif args.model in ['resnet50_v1', 'resnet101_v1']:
+    elif args.model in ['resnet18_v1', 'resnet50_v1', 'resnet101_v1']:
         rgb_mean = '123.68,116.779,103.939'
         rgb_std = '58.393, 57.12, 57.375'
         if exclude_first_conv:
@@ -234,6 +242,12 @@ def save_params(fname, arg_params, aux_params, logger=None):
                                'mobilenet0_pool0_fwd']
         if exclude_first_conv:
             excluded_sym_names += ['mobilenet0_conv0_fwd']
+    elif args.model == 'mobilenetv2_1.0':
+        rgb_mean = '123.68,116.779,103.939'
+        rgb_std = '58.393, 57.12, 57.375'
+        excluded_sym_names += ['mobilenetv20_output_flatten0_flatten0']
+        if exclude_first_conv:
+            excluded_sym_names += ['mobilenetv20_conv0_fwd']
     elif args.model == 'inceptionv3':
         rgb_mean = '123.68,116.779,103.939'
         rgb_std = '58.393, 57.12, 57.375'
diff --git a/src/operator/nn/mkldnn/mkldnn_base.cc b/src/operator/nn/mkldnn/mkldnn_base.cc
index e80358ac636a..5dccba281fd0 100644
--- a/src/operator/nn/mkldnn/mkldnn_base.cc
+++ b/src/operator/nn/mkldnn/mkldnn_base.cc
@@ -392,6 +392,7 @@ mkldnn_memory_format_t GetDefaultFormat(const mkldnn::memory::desc &desc) {
       case mkldnn_gOhwi8o:
       case mkldnn_gOhwi16o:
       case mkldnn_gOhIw16o4i:
+      case mkldnn_Goihw16g_s8s8:
         return mkldnn_goihw;
       default:
         LOG(FATAL) << "Unknown MKLDNN format for 5 dimensions: " << desc.data.format;