Add vgg16 & resnet50 to supported models

[kuke]

VGG16 Validation. The size of two intermediate fully-connected layers is set to 512, for the case which is set to 4096, the inference of ONNX model gets too slow.

python validate.py --fluid_model image_classification_vgg_512.inference.model/ --onnx_model image_classification_vgg_512.onnx 

----------- Configuration Arguments ----------- a: 0.0 b: 1.0 batch_size: 10 expected_decimal: 5 fluid_model: image_classification_vgg_512.inference.model/ onnx_model: image_classification_vgg_512.onnx ------------------------------------------------ Inference results for fluid model: [array([[0.10089886, 0.09871482, 0.10326847, 0.10127534, 0.09726173, 0.10202399, 0.09696591, 0.09893821, 0.10302259, 0.09763013], [0.10091323, 0.09872197, 0.10325751, 0.10122918, 0.09734982, 0.10196733, 0.09698458, 0.09890091, 0.10304806, 0.09762745], [0.10088442, 0.09874254, 0.10324674, 0.10130584, 0.09730507, 0.10196551, 0.09692121, 0.09891873, 0.10305765, 0.09765223], [0.1009225 , 0.09880089, 0.10329139, 0.10122177, 0.09734076, 0.1019473 , 0.09699273, 0.09891348, 0.10295378, 0.09761531], [0.10091846, 0.09871273, 0.10328399, 0.10130198, 0.09726519, 0.10197122, 0.09687649, 0.09894703, 0.1030934 , 0.09762951], [0.10092277, 0.09875839, 0.1032907 , 0.10128019, 0.0973415 , 0.10200986, 0.09692244, 0.09887472, 0.10299625, 0.09760312], [0.10092793, 0.09872951, 0.10329424, 0.1012623 , 0.09729295, 0.10197538, 0.09699677, 0.0988729 , 0.10303052, 0.09761748], [0.10088892, 0.09871186, 0.10328761, 0.10128257, 0.09727918, 0.10195947, 0.09692416, 0.09894343, 0.10303068, 0.09769206], [0.10086466, 0.09880146, 0.10325968, 0.10126632, 0.09731811, 0.10198917, 0.09694263, 0.09888101, 0.10306235, 0.09761455], [0.10084924, 0.09872525, 0.10329291, 0.10123692, 0.09729046, 0.10201162, 0.096938 , 0.09895328, 0.10304983, 0.09765254]], dtype=float32)] Inference results for ONNX model: Outputs(_0=array([[0.10089886, 0.09871481, 0.10326847, 0.10127534, 0.09726173, 0.10202398, 0.09696591, 0.09893821, 0.10302258, 0.09763013], [0.10091322, 0.09872196, 0.10325751, 0.10122917, 0.09734982, 0.10196733, 0.09698458, 0.0989009 , 0.10304805, 0.09762745], [0.10088442, 0.09874254, 0.10324674, 0.10130584, 0.09730507, 0.10196552, 0.09692121, 0.09891874, 0.10305766, 0.09765223], [0.1009225 , 0.09880089, 0.10329139, 0.10122177, 0.09734076, 0.1019473 , 0.09699273, 0.09891348, 0.10295378, 0.09761531], [0.10091845, 0.09871273, 0.10328399, 0.10130197, 0.09726518, 0.1019712 , 0.09687648, 0.09894701, 0.10309339, 0.09762949], [0.10092278, 0.0987584 , 0.10329071, 0.1012802 , 0.09734151, 0.10200986, 0.09692245, 0.09887473, 0.10299627, 0.09760312], [0.10092794, 0.09872951, 0.10329424, 0.1012623 , 0.09729295, 0.10197538, 0.09699677, 0.0988729 , 0.10303052, 0.09761748], [0.10088893, 0.09871187, 0.10328761, 0.10128257, 0.09727918, 0.10195947, 0.09692415, 0.09894343, 0.10303069, 0.09769207], [0.10086467, 0.09880148, 0.10325969, 0.10126634, 0.09731812, 0.10198919, 0.09694266, 0.09888103, 0.10306237, 0.09761456], [0.10084924, 0.09872525, 0.10329291, 0.10123691, 0.09729046, 0.10201162, 0.09693799, 0.09895328, 0.10304983, 0.09765254]], dtype=float32)) The exported model achieves 5-decimal precision. 

[kuke]

ResNet50

python validate.py --fluid_model extras/image_classification_resnet.inference.model/ --onnx_model image_classification_resnet.onnx 

----------- Configuration Arguments ----------- a: 0.0 b: 1.0 batch_size: 10 expected_decimal: 5 fluid_model: extras/image_classification_resnet.inference.model/ onnx_model: image_classification_resnet.onnx ------------------------------------------------ Inference results for fluid model: [array([[1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28], [1.6038109e-28, 1.0000000e+00, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28, 1.6038109e-28]], dtype=float32)] Inference results for ONNX model: Outputs(_0=array([[0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.]], dtype=float32)) The exported model achieves 5-decimal precision. 

[varunarora]

interesting! why is it slow?

[kuke]

@varunarora I haven't had a detailed analysis. I think we'd better profile the network and find the bottleneck if we care about computation performance someday.