This folder contains building code for MobileNetV2, based on MobileNetV2: Inverted Residuals and Linear Bottlenecks
This model file has been pushed to my keras fork which you can see here. You can also view the active pull request to keras here A lot of the techniques and strategies developed for weight extraction in this repository was taken from here
This is the timing of MobileNetV1 vs MobileNetV2 using TF-Lite on the large core of Pixel 1 phone.
| mobilenet_v2_1.0_224 | 300 | 3.47 | 71.8 | 91.0 | 73.8
First, I chose to extract all the weights from Tensorflows repo and save the depth_multiplier values and input resolutions to a file models_to_load.py.
The for each model in models_to_load.py, I extracted the weights using file extract_weights.py, utilizing the checkpoints provided, and saved the weights to a directory called 'weights'. Then I used the file load_weights_multiple.py to set the weights of the corresponding keras model using keras's built in set_weights function. I used a pickle file that was generated using extract_weights.py to serve as a guide and provide meta data about each layer so that I could align them. Each weight is checked for:
Shape, mod (expand, depthwise, or project), meta: (weights or batch norm parameters), and size.
The model is then tested inside test_mobilenet.py. This model is tested against the tensorflow slim model that can be found here
to use this model:
from keras.applications.mobilenetv2 import MobileNetV2
from keras.layers import Input
input_tensor = Input(shape=(224,224, 3)) # or you could put (None, None, 3) for shape
model = MobileNetV2(input_tensor = input_tensor, alpha = 1.0, include_top = True, weights=’imagenet’)
# Now you have a fully loaded model.
The model only works with depth_multiplier = 1, although the alpha parameter is able to specify width_multipliers if they are included in [0.35, 0.50, 0.75, 1.0] Additionally, only square input sizes included in [96, 128, 160, 182, 224] can be used.
The include_top parameter can be used to grab the full network, if you set it to false, you will grab the base network before the pooling operation and fully connected layer.
Models can be found here
- These results are taken from tfmobilenet but I estimate ours are similar in performance. Except for the Pixel 1 inference time.
You can grab and load up the pickle file test_results.p or you can read the results below: Please note that there are subtle differences between the TF models and the Keras models in the testing procedure, these are due to
the differences in how Keras performs softmax, and the normalization that occurs after we pop out the first tensorflow logit (that is the background class) and re-normalize.
For questions, comments, and concerns please reach me at [email protected].
test_results: [{
'rows': 224,
'vector_difference': array([
[3.06442744e-05, 1.03940765e-05, 6.52904509e-06, ...,
1.86086560e-04, 1.36749877e-05, 2.29768884e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.4,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 0.584200382232666,
'inference_time_keras': 2.468465566635132,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.4_224.h5',
'max_vector_difference': 0.0442425,
'preds_agree': True
}, {
'rows': 224,
'vector_difference': array([
[1.2600726e-04, 1.9243037e-04, 8.6632121e-05, ..., 1.7771832e-05,
1.2540509e-04, 8.6921602e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.3,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 0.7864320278167725,
'inference_time_keras': 0.48574304580688477,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.3_224.h5',
'max_vector_difference': 0.1639086,
'preds_agree': True
}, {
'rows': 224,
'vector_difference': array([
[2.6156631e-06, 8.9272799e-06, 9.9282261e-07, ..., 2.7298967e-05,
7.0550705e-06, 1.7008846e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.0,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 0.9191036224365234,
'inference_time_keras': 0.5298285484313965,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_224.h5',
'max_vector_difference': 0.008201845,
'preds_agree': True
}, {
'rows': 192,
'vector_difference': array([
[2.9478622e-05, 3.1182157e-05, 1.6525744e-05, ..., 3.2548003e-05,
2.3264165e-05, 1.2547210e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.0,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 1.0997955799102783,
'inference_time_keras': 0.609818696975708,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_192.h5',
'max_vector_difference': 0.03410864,
'preds_agree': True
}, {
'rows': 160,
'vector_difference': array([
[1.9924228e-06, 1.1813272e-05, 2.6646510e-05, ..., 2.1615942e-06,
7.3942429e-06, 2.8581535e-06
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.0,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 1.3430328369140625,
'inference_time_keras': 0.6801567077636719,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_160.h5',
'max_vector_difference': 0.0024641072,
'preds_agree': True
}, {
'rows': 128,
'vector_difference': array([
[1.9651561e-05, 7.6118726e-05, 1.1588483e-05, ..., 2.3098060e-05,
5.0026181e-05, 2.3178840e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.0,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 1.6118056774139404,
'inference_time_keras': 0.7277970314025879,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_128.h5',
'max_vector_difference': 0.022542655,
'preds_agree': True
}, {
'rows': 96,
'vector_difference': array([
[4.1969906e-07, 6.2867985e-06, 7.6682009e-06, ..., 9.5812502e-06,
5.6552781e-06, 2.0793846e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 1.0,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 1.759774923324585,
'inference_time_keras': 0.8093435764312744,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_96.h5',
'max_vector_difference': 0.009057283,
'preds_agree': True
}, {
'rows': 224,
'vector_difference': array([
[1.9891046e-05, 3.5158137e-05, 7.4801164e-06, ..., 3.1968251e-05,
1.8171089e-05, 2.4247890e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.75,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 1.988135576248169,
'inference_time_keras': 0.907604455947876,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_224.h5',
'max_vector_difference': 0.027664006,
'preds_agree': True
}, {
'rows': 192,
'vector_difference': array([
[2.4571000e-06, 2.3607154e-06, 1.0745002e-06, ..., 7.6524229e-06,
2.5452073e-07, 6.0848397e-06
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.75,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 2.1681172847747803,
'inference_time_keras': 0.9792499542236328,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_192.h5',
'max_vector_difference': 0.0048509836,
'preds_agree': True
}, {
'rows': 160,
'vector_difference': array([
[3.3421951e-05, 2.8039271e-05, 2.6024140e-05, ..., 1.8016202e-05,
2.8351524e-06, 3.2267882e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.75,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 2.4570868015289307,
'inference_time_keras': 1.0636296272277832,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_160.h5',
'max_vector_difference': 0.04817468,
'preds_agree': True
}, {
'rows': 128,
'vector_difference': array([
[5.7490397e-05, 2.3515218e-05, 6.2765699e-05, ..., 6.3877407e-05,
2.4530049e-05, 1.9020826e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.75,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 2.712172746658325,
'inference_time_keras': 1.1841137409210205,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_128.h5',
'max_vector_difference': 0.10443801,
'preds_agree': True
}, {
'rows': 96,
'vector_difference': array([
[1.91572035e-05, 1.15800685e-05, 4.44647230e-06, ...,
9.32329567e-06, 4.12614900e-05, 3.88475746e-06
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.75,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 2.947575807571411,
'inference_time_keras': 1.2835781574249268,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_96.h5',
'max_vector_difference': 0.02189435,
'preds_agree': True
}, {
'rows': 224,
'vector_difference': array([
[2.7594273e-05, 1.8192208e-05, 5.0051967e-06, ..., 2.2260952e-05,
1.0851298e-05, 2.0575267e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.5,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 3.234971761703491,
'inference_time_keras': 1.4089748859405518,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_224.h5',
'max_vector_difference': 0.03057003,
'preds_agree': True
}, {
'rows': 192,
'vector_difference': array([
[6.4921463e-05, 1.8974228e-05, 9.5047453e-06, ..., 2.7531139e-05,
1.5725660e-05, 1.3637640e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.5,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 3.471426010131836,
'inference_time_keras': 1.5226759910583496,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_192.h5',
'max_vector_difference': 0.10459131,
'preds_agree': True
}, {
'rows': 160,
'vector_difference': array([
[2.1799133e-05, 2.6465546e-05, 9.7673910e-07, ..., 4.9333670e-05,
1.2139077e-05, 3.4930854e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.5,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 3.7173619270324707,
'inference_time_keras': 1.6358251571655273,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_160.h5',
'max_vector_difference': 0.041921377,
'preds_agree': True
}, {
'rows': 128,
'vector_difference': array([
[1.6039543e-05, 3.6521582e-05, 2.0016232e-06, ..., 7.7442382e-06,
1.3480414e-05, 1.9661791e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.5,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 3.96859073638916,
'inference_time_keras': 1.7203476428985596,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_128.h5',
'max_vector_difference': 0.015232503,
'preds_agree': True
}, {
'rows': 96,
'vector_difference': array([
[5.3964366e-05, 4.3542765e-05, 1.9309173e-05, ..., 1.8680606e-05,
1.7692482e-05, 1.0907562e-03
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.5,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 4.313321590423584,
'inference_time_keras': 1.8665125370025635,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_96.h5',
'max_vector_difference': 0.10974246,
'preds_agree': True
}, {
'rows': 224,
'vector_difference': array([
[1.8852079e-05, 9.4596544e-07, 2.4118672e-06, ..., 3.3465330e-06,
3.7501377e-07, 2.0572159e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.35,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 4.690372705459595,
'inference_time_keras': 1.9682881832122803,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_224.h5',
'max_vector_difference': 0.018127322,
'preds_agree': True
}, {
'rows': 192,
'vector_difference': array([
[5.89300471e-05, 1.10333494e-05, 2.08540587e-06, ...,
1.03199854e-04, 1.02247395e-05, 3.36650060e-04
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.35,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 4.746210336685181,
'inference_time_keras': 2.11893892288208,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_192.h5',
'max_vector_difference': 0.031285435,
'preds_agree': True
}, {
'rows': 160,
'vector_difference': array([
[4.2083520e-06, 5.5578494e-07, 3.2600292e-07, ..., 4.2584943e-06,
5.7042635e-06, 4.6083354e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.35,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 5.078217029571533,
'inference_time_keras': 2.330106735229492,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_160.h5',
'max_vector_difference': 0.008356452,
'preds_agree': True
}, {
'rows': 128,
'vector_difference': array([
[1.0704320e-05, 5.3489948e-06, 7.2533185e-06, ..., 1.6965925e-05,
2.3451194e-06, 2.8069786e-05
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.35,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 5.38210391998291,
'inference_time_keras': 2.5224623680114746,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_128.h5',
'max_vector_difference': 0.012936056,
'preds_agree': True
}, {
'rows': 96,
'vector_difference': array([
[6.1693572e-06, 3.9814022e-06, 3.0157253e-07, ..., 4.4240751e-06,
2.7236201e-06, 7.0944225e-06
]
], dtype = float32),
'pred_keras_score': 389,
'alpha': 0.35,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_tf_score': 389,
'inference_time_tf': 5.456079721450806,
'inference_time_keras': 2.454572916030884,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_96.h5',
'max_vector_difference': 0.018446982,
'preds_agree': True
}]
[{
'rows': 224,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.4,
'inference_time_keras': 0.2321312427520752,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropodamelanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.4_224.h5',
'max_vector_difference': 0.042831063,
'inference_time_tf': 0.5365610122680664
}, {
'rows': 224,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.3,
'inference_time_keras': 0.309173583984375,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.3_224.h5',
'max_vector_difference': 0.1633901,
'inference_time_tf': 0.6658704280853271
}, {
'rows': 224,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.0,
'inference_time_keras': 0.36829209327697754,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_224.h5',
'max_vector_difference': 0.00815512,
'inference_time_tf': 0.8123471736907959
}, {
'rows': 192,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.0,
'inference_time_keras': 0.43697261810302734,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_192.h5',
'max_vector_difference': 0.03405291,
'inference_time_tf': 1.0188896656036377
}, {
'rows': 160,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.0,
'inference_time_keras': 0.5213620662689209,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_160.h5',
'max_vector_difference': 0.0024634155,
'inference_time_tf': 1.231605052947998
}, {
'rows': 128,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.0,
'inference_time_keras': 0.590782642364502,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_128.h5',
'max_vector_difference': 0.02254337,
'inference_time_tf': 1.4477722644805908
}, {
'rows': 96,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 1.0,
'inference_time_keras': 0.6745755672454834,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_96.h5',
'max_vector_difference': 0.009703398,
'inference_time_tf': 1.725059986114502
}, {
'rows': 224,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.75,
'inference_time_keras': 0.7602894306182861,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_224.h5',
'max_vector_difference': 0.03044498,
'inference_time_tf': 1.8605682849884033
}, {
'rows': 192,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.75,
'inference_time_keras': 0.8398780822753906,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_192.h5',
'max_vector_difference': 0.0047159195,
'inference_time_tf': 2.006946086883545
}, {
'rows': 160,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.75,
'inference_time_keras': 0.9440453052520752,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_160.h5',
'max_vector_difference': 0.048369706,
'inference_time_tf': 2.309004545211792
}, {
'rows': 128,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.75,
'inference_time_keras': 1.0247721672058105,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_128.h5',
'max_vector_difference': 0.10489011,
'inference_time_tf': 2.551898717880249
}, {
'rows': 96,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.75,
'inference_time_keras': 1.1196868419647217,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.75_96.h5',
'max_vector_difference': 0.020857051,
'inference_time_tf': 2.7527146339416504
}, {
'rows': 224,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.5,
'inference_time_keras': 1.230698823928833,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_224.h5',
'max_vector_difference': 0.031421363,
'inference_time_tf': 2.9628419876098633
}, {
'rows': 192,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.5,
'inference_time_keras': 1.331193447113037,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_192.h5',
'max_vector_difference': 0.10464072,
'inference_time_tf': 3.2052600383758545
}, {
'rows': 160,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.5,
'inference_time_keras': 1.4506702423095703,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_160.h5',
'max_vector_difference': 0.04293281,
'inference_time_tf': 3.4141736030578613
}, {
'rows': 128,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.5,
'inference_time_keras': 1.546645164489746,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_128.h5',
'max_vector_difference': 0.012662649,
'inference_time_tf': 3.6236307621002197
}, {
'rows': 96,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.5,
'inference_time_keras': 1.6599667072296143,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.5_96.h5',
'max_vector_difference': 0.11007625,
'inference_time_tf': 3.9398250579833984
}, {
'rows': 224,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.35,
'inference_time_keras': 1.7787129878997803,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_224.h5',
'max_vector_difference': 0.018119395,
'inference_time_tf': 4.164514064788818
}, {
'rows': 192,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.35,
'inference_time_keras': 1.8967313766479492,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_192.h5',
'max_vector_difference': 0.031832576,
'inference_time_tf': 4.384629249572754
}, {
'rows': 160,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.35,
'inference_time_keras': 2.034467935562134,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_160.h5',
'max_vector_difference': 0.008257747,
'inference_time_tf': 4.6021528244018555
}, {
'rows': 128,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.35,
'inference_time_keras': 2.175736427307129,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_128.h5',
'max_vector_difference': 0.012835741,
'inference_time_tf': 4.958279609680176
}, {
'rows': 96,
'pred_tf_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'pred_keras_score': 389,
'pred_tf_score': 389,
'preds_agree': True,
'alpha': 0.35,
'inference_time_keras': 2.609691619873047,
'pred_keras_label': 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
'model': '/home/jon/Documents/keras_mobilenetV2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_96.h5',
'max_vector_difference': 0.019455373,
'inference_time_tf': 6.363346576690674
}]