Note
Click here to download the full example code
1. Export trained GluonCV network to JSON¶Download Python source code: export_network.pyDownload Jupyter notebook: export_network.ipynb
It is awesome if you are enjoy using GluonCV in Python for training and testing. At some point, you might ask: “Is it possible to deploy the existing models to somewhere out of Python environments?”
The answer is “Absolutely!”, and it’s super easy actually.
This article will show you how to export networks/models to be used somewhere other than Python.
import gluoncv as gcv
from gluoncv.utils import export_block
First of all, we need a network to play with, a pre-trained one is perfect
net = gcv.model_zoo.get_model('resnet18_v1', pretrained=True)
export_block('resnet18_v1', net, preprocess=True, layout='HWC')
print('Done.')
Out:
Downloading /root/.mxnet/models/resnet18_v1-a0666292.zip from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/resnet18_v1-a0666292.zip...
0%| | 0/42434 [00:00<?, ?KB/s]
0%| | 100/42434 [00:00<00:52, 799.68KB/s]
1%|1 | 506/42434 [00:00<00:18, 2259.47KB/s]
4%|4 | 1753/42434 [00:00<00:07, 5803.45KB/s]
17%|#6 | 7040/42434 [00:00<00:01, 21999.06KB/s]
31%|### | 13017/42434 [00:00<00:00, 34445.25KB/s]
50%|##### | 21256/42434 [00:00<00:00, 47811.17KB/s]
67%|######7 | 28544/42434 [00:00<00:00, 55457.95KB/s]
83%|########3 | 35368/42434 [00:00<00:00, 59337.18KB/s]
42435KB [00:00, 43650.32KB/s]
Done.
Hint
Use preprocess=True will add a default preprocess layer above the network,
which will subtract mean [123.675, 116.28, 103.53], divide
std [58.395, 57.12, 57.375], and convert original image (B, H, W, C and range [0, 255]) to
tensor (B, C, H, W) as network input. This is the default preprocess behavior of all GluonCV
pre-trained models. With this preprocess head, you can use raw RGB image in C++ without
explicitly applying these operations.
The above code generates two files: xxxx.json and xxxx.params
Out:
['resnet18_v1-symbol.json', 'resnet18_v1-0000.params']
JSON file includes computational graph and params file includes pre-trained weights.
The exportable networks are not limited to image classification models. We can export detection and segmentation networks as well:
# YOLO
net = gcv.model_zoo.get_model('yolo3_darknet53_coco', pretrained=True)
export_block('yolo3_darknet53_coco', net)
# FCN
net = gcv.model_zoo.get_model('fcn_resnet50_ade', pretrained=True)
export_block('fcn_resnet50_ade', net)
# MaskRCNN
net = gcv.model_zoo.get_model('mask_rcnn_resnet50_v1b_coco', pretrained=True)
export_block('mask_rcnn_resnet50_v1b_coco', net)
Out:
Downloading /root/.mxnet/models/yolo3_darknet53_coco-09767802.zip from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/yolo3_darknet53_coco-09767802.zip...
0%| | 0/224190 [00:00<?, ?KB/s]
0%| | 102/224190 [00:00<04:40, 797.94KB/s]
0%| | 508/224190 [00:00<01:42, 2189.18KB/s]
1%| | 2184/224190 [00:00<00:30, 7162.90KB/s]
3%|3 | 7758/224190 [00:00<00:09, 23548.98KB/s]
7%|6 | 14640/224190 [00:00<00:05, 38269.95KB/s]
10%|9 | 22248/224190 [00:00<00:04, 50209.52KB/s]
13%|#3 | 30133/224190 [00:00<00:03, 59133.22KB/s]
17%|#6 | 37825/224190 [00:00<00:02, 64607.58KB/s]
20%|## | 45698/224190 [00:00<00:02, 68922.24KB/s]
24%|##3 | 53787/224190 [00:01<00:02, 72556.99KB/s]
27%|##7 | 61507/224190 [00:01<00:02, 73959.48KB/s]
31%|### | 69308/224190 [00:01<00:02, 75180.34KB/s]
34%|###4 | 76889/224190 [00:01<00:01, 75366.99KB/s]
38%|###7 | 84792/224190 [00:01<00:01, 76465.43KB/s]
41%|####1 | 92776/224190 [00:01<00:01, 77447.89KB/s]
45%|####4 | 100539/224190 [00:01<00:01, 76307.41KB/s]
49%|####8 | 108984/224190 [00:01<00:01, 78723.51KB/s]
52%|#####2 | 116871/224190 [00:01<00:01, 77722.21KB/s]
56%|#####5 | 124968/224190 [00:01<00:01, 78389.09KB/s]
59%|#####9 | 132816/224190 [00:02<00:01, 78382.50KB/s]
63%|######2 | 140661/224190 [00:02<00:01, 76579.59KB/s]
66%|######6 | 148332/224190 [00:02<00:01, 69287.50KB/s]
70%|######9 | 156747/224190 [00:02<00:00, 73369.48KB/s]
73%|#######3 | 164309/224190 [00:02<00:00, 74002.97KB/s]
77%|#######7 | 172629/224190 [00:02<00:00, 76640.69KB/s]
80%|######## | 180372/224190 [00:02<00:00, 76389.55KB/s]
84%|########3 | 188066/224190 [00:02<00:00, 76300.68KB/s]
88%|########7 | 196517/224190 [00:02<00:00, 78709.86KB/s]
91%|#########1| 204420/224190 [00:03<00:00, 77343.70KB/s]
95%|#########4| 212181/224190 [00:03<00:00, 77060.98KB/s]
98%|#########8| 220169/224190 [00:03<00:00, 77880.92KB/s]
100%|##########| 224190/224190 [00:03<00:00, 67850.24KB/s]
Downloading /root/.mxnet/models/fcn_resnet50_ade-3479525a.zip from https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/models/fcn_resnet50_ade-3479525a.zip...
0%| | 0/128831 [00:00<?, ?KB/s]
0%| | 100/128831 [00:00<02:44, 780.55KB/s]
0%| | 505/128831 [00:00<00:59, 2169.61KB/s]
2%|1 | 2177/128831 [00:00<00:18, 6790.34KB/s]
6%|5 | 7385/128831 [00:00<00:05, 21559.64KB/s]
11%|# | 13635/128831 [00:00<00:03, 34569.89KB/s]
17%|#6 | 21329/128831 [00:00<00:02, 46431.54KB/s]
23%|##2 | 29117/128831 [00:00<00:01, 55934.25KB/s]
28%|##8 | 36303/128831 [00:00<00:01, 60735.86KB/s]
35%|###4 | 44870/128831 [00:01<00:01, 68241.33KB/s]
40%|#### | 51863/128831 [00:01<00:01, 68202.32KB/s]
46%|####5 | 59161/128831 [00:01<00:01, 69625.84KB/s]
52%|#####1 | 66497/128831 [00:01<00:00, 70737.68KB/s]
58%|#####8 | 74946/128831 [00:01<00:00, 74853.17KB/s]
64%|######4 | 82479/128831 [00:01<00:00, 73963.49KB/s]
70%|######9 | 90007/128831 [00:01<00:00, 74268.49KB/s]
76%|#######5 | 97709/128831 [00:01<00:00, 75084.79KB/s]
82%|########1 | 105481/128831 [00:01<00:00, 75869.74KB/s]
88%|########7 | 113082/128831 [00:01<00:00, 75685.34KB/s]
94%|#########4| 121217/128831 [00:02<00:00, 77214.22KB/s]
128832KB [00:02, 60766.94KB/s]
We are all set here. Please checkout the other tutorials of how to use the JSON and params files.
Total running time of the script: ( 0 minutes 22.950 seconds)