Mask R-CNN - Inspect Weights of a Trained Model

This notebook includes code and visualizations to test, debug, and evaluate the Mask R-CNN model.

import os
import sys
import numpy as np
import tensorflow as tf
import matplotlib
import matplotlib.pyplot as plt
import keras

# Root directory of the project
ROOT_DIR = os.path.abspath("../../")

# Import Mask RCNN
sys.path.append(ROOT_DIR)  # To find local version of the library
from mrcnn import utils
import mrcnn.model as modellib
from mrcnn import visualize
from mrcnn.model import log

%matplotlib inline 

# Directory to save logs and trained model
MODEL_DIR = os.path.join(ROOT_DIR, "logs")

# Local path to trained weights file
COCO_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_coco.h5")
# Download COCO trained weights from Releases if needed
if not os.path.exists(COCO_MODEL_PATH):
    utils.download_trained_weights(COCO_MODEL_PATH)

# Path to Shapes trained weights
SHAPES_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_shapes.h5")

Using TensorFlow backend.

Configurations

# Run one of the code blocks

# Shapes toy dataset
# import shapes
# config = shapes.ShapesConfig()

# MS COCO Dataset
import coco
config = coco.CocoConfig()

Notebook Preferences

# Device to load the neural network on.
# Useful if you're training a model on the same 
# machine, in which case use CPU and leave the
# GPU for training.
DEVICE = "/cpu:0"  # /cpu:0 or /gpu:0

def get_ax(rows=1, cols=1, size=16):
    """Return a Matplotlib Axes array to be used in
    all visualizations in the notebook. Provide a
    central point to control graph sizes.
    
    Adjust the size attribute to control how big to render images
    """
    _, ax = plt.subplots(rows, cols, figsize=(size*cols, size*rows))
    return ax

Load Model

# Create model in inference mode
with tf.device(DEVICE):
    model = modellib.MaskRCNN(mode="inference", model_dir=MODEL_DIR,
                              config=config)

# Set weights file path
if config.NAME == "shapes":
    weights_path = SHAPES_MODEL_PATH
elif config.NAME == "coco":
    weights_path = COCO_MODEL_PATH
# Or, uncomment to load the last model you trained
# weights_path = model.find_last()

# Load weights
print("Loading weights ", weights_path)
model.load_weights(weights_path, by_name=True)

Review Weight Stats

# Show stats of all trainable weights    
visualize.display_weight_stats(model)

conv1/kernel:0	(7, 7, 3, 64)	-0.8616	+0.8539	+0.1314
bn_conv1/gamma:0	(64,)	+0.0843	+2.6420	+0.5087
bn_conv1/beta:0	(64,)	-2.4174	+5.4189	+1.9981
bn_conv1/moving_mean:0	(64,)	-172.9685	+94.5717	+42.0063
bn_conv1/moving_variance:0*** Overflow?	(64,)	+0.0000	+110557.9688	+16228.7607
res2a_branch2a/kernel:0	(1, 1, 64, 64)	-0.6603	+0.3208	+0.0768
bn2a_branch2a/gamma:0	(64,)	+0.2189	+1.8654	+0.4149
bn2a_branch2a/beta:0	(64,)	-2.1375	+3.7690	+1.1904
bn2a_branch2a/moving_mean:0	(64,)	-6.3118	+7.4370	+2.4037
bn2a_branch2a/moving_variance:0	(64,)	+0.0000	+8.8091	+2.1498
res2a_branch2b/kernel:0	(3, 3, 64, 64)	-0.3813	+0.5123	+0.0323
bn2a_branch2b/gamma:0	(64,)	+0.3195	+1.7454	+0.3143
bn2a_branch2b/beta:0	(64,)	-1.9530	+4.5882	+1.5261
bn2a_branch2b/moving_mean:0	(64,)	-6.7890	+4.2754	+2.2064
bn2a_branch2b/moving_variance:0	(64,)	+0.0000	+5.5464	+1.1573
res2a_branch2c/kernel:0	(1, 1, 64, 256)	-0.4412	+0.3600	+0.0411
res2a_branch1/kernel:0	(1, 1, 64, 256)	-0.8513	+0.7543	+0.0699
bn2a_branch2c/gamma:0	(256,)	-0.5887	+3.2101	+0.6259
bn2a_branch2c/beta:0	(256,)	-1.1511	+1.4415	+0.4269
bn2a_branch2c/moving_mean:0	(256,)	-4.2796	+3.1055	+1.0352
bn2a_branch2c/moving_variance:0	(256,)	+0.0000	+2.6966	+0.4085
bn2a_branch1/gamma:0	(256,)	+0.2415	+3.5354	+0.6298
bn2a_branch1/beta:0	(256,)	-1.1511	+1.4415	+0.4269
bn2a_branch1/moving_mean:0	(256,)	-8.1191	+8.7749	+2.0398
bn2a_branch1/moving_variance:0	(256,)	+0.0000	+10.3201	+1.6540
res2b_branch2a/kernel:0	(1, 1, 256, 64)	-0.2418	+0.2263	+0.0358
bn2b_branch2a/gamma:0	(64,)	+0.2051	+1.7890	+0.3852
bn2b_branch2a/beta:0	(64,)	-2.0730	+1.6836	+0.8930
bn2b_branch2a/moving_mean:0	(64,)	-1.8157	+1.7829	+0.7466
bn2b_branch2a/moving_variance:0	(64,)	+0.0000	+3.2496	+0.7830
res2b_branch2b/kernel:0	(3, 3, 64, 64)	-0.5190	+0.3431	+0.0357
bn2b_branch2b/gamma:0	(64,)	+0.5190	+1.4828	+0.2283
bn2b_branch2b/beta:0	(64,)	-2.4756	+2.7818	+1.2069
bn2b_branch2b/moving_mean:0	(64,)	-1.8361	+0.9368	+0.5723
bn2b_branch2b/moving_variance:0	(64,)	+0.0938	+1.0783	+0.2077
res2b_branch2c/kernel:0	(1, 1, 64, 256)	-0.3330	+0.3228	+0.0414
bn2b_branch2c/gamma:0	(256,)	-0.0329	+1.8095	+0.4257
bn2b_branch2c/beta:0	(256,)	-1.3059	+0.9721	+0.3463
bn2b_branch2c/moving_mean:0	(256,)	-2.5336	+2.1111	+0.5033
bn2b_branch2c/moving_variance:0	(256,)	+0.0000	+0.2187	+0.0333
res2c_branch2a/kernel:0	(1, 1, 256, 64)	-0.3040	+0.2175	+0.0412
bn2c_branch2a/gamma:0	(64,)	+0.2683	+1.8338	+0.2863
bn2c_branch2a/beta:0	(64,)	-2.0358	+0.8512	+0.7946
bn2c_branch2a/moving_mean:0	(64,)	-4.7340	+1.6664	+1.2255
bn2c_branch2a/moving_variance:0	(64,)	+0.0000	+3.4985	+0.7644
res2c_branch2b/kernel:0	(3, 3, 64, 64)	-0.2020	+0.2138	+0.0378
bn2c_branch2b/gamma:0	(64,)	+0.6155	+1.5482	+0.2177
bn2c_branch2b/beta:0	(64,)	-2.4321	+1.8318	+0.6352
bn2c_branch2b/moving_mean:0	(64,)	-1.4939	+0.1422	+0.2628
bn2c_branch2b/moving_variance:0	(64,)	+0.2278	+2.0831	+0.3304
res2c_branch2c/kernel:0	(1, 1, 64, 256)	-0.2842	+0.2529	+0.0430
bn2c_branch2c/gamma:0	(256,)	-0.0200	+2.3871	+0.5297
bn2c_branch2c/beta:0	(256,)	-1.6989	+1.1085	+0.4321
bn2c_branch2c/moving_mean:0	(256,)	-1.2794	+0.7256	+0.2929
bn2c_branch2c/moving_variance:0	(256,)	+0.0010	+0.7414	+0.1125
res3a_branch2a/kernel:0	(1, 1, 256, 128)	-0.5027	+0.6187	+0.0304
bn3a_branch2a/gamma:0	(128,)	+0.4905	+1.3262	+0.1895
bn3a_branch2a/beta:0	(128,)	-1.8565	+2.5853	+0.7641
bn3a_branch2a/moving_mean:0	(128,)	-4.2267	+2.1703	+0.9005
bn3a_branch2a/moving_variance:0	(128,)	+0.0545	+8.9011	+1.2696
res3a_branch2b/kernel:0	(3, 3, 128, 128)	-0.3236	+0.4518	+0.0223
bn3a_branch2b/gamma:0	(128,)	+0.4706	+1.8430	+0.2200
bn3a_branch2b/beta:0	(128,)	-1.9615	+1.9157	+0.7933
bn3a_branch2b/moving_mean:0	(128,)	-6.0335	+3.2213	+1.9778
bn3a_branch2b/moving_variance:0	(128,)	+0.0001	+6.6136	+0.8450
res3a_branch2c/kernel:0	(1, 1, 128, 512)	-0.4927	+0.3402	+0.0283
res3a_branch1/kernel:0	(1, 1, 256, 512)	-0.4507	+0.6643	+0.0290
bn3a_branch2c/gamma:0	(512,)	-0.0033	+3.7310	+0.6223
bn3a_branch2c/beta:0	(512,)	-0.9694	+1.4581	+0.3727
bn3a_branch2c/moving_mean:0	(512,)	-1.5891	+1.4301	+0.3907
bn3a_branch2c/moving_variance:0	(512,)	+0.0002	+0.8632	+0.1073
bn3a_branch1/gamma:0	(512,)	-0.0138	+2.7207	+0.4812
bn3a_branch1/beta:0	(512,)	-0.9694	+1.4580	+0.3727
bn3a_branch1/moving_mean:0	(512,)	-3.7382	+2.8448	+0.8043
bn3a_branch1/moving_variance:0	(512,)	+0.0029	+5.7492	+0.6063
res3b_branch2a/kernel:0	(1, 1, 512, 128)	-0.1992	+0.1954	+0.0252
bn3b_branch2a/gamma:0	(128,)	+0.5946	+1.5472	+0.1807
bn3b_branch2a/beta:0	(128,)	-3.9918	+0.6877	+0.6492
bn3b_branch2a/moving_mean:0	(128,)	-3.0865	+1.0082	+0.6388
bn3b_branch2a/moving_variance:0	(128,)	+0.2999	+3.6467	+0.6298
res3b_branch2b/kernel:0	(3, 3, 128, 128)	-0.2305	+0.2760	+0.0241
bn3b_branch2b/gamma:0	(128,)	+0.4863	+1.5054	+0.2365
bn3b_branch2b/beta:0	(128,)	-2.4436	+1.4355	+0.6833
bn3b_branch2b/moving_mean:0	(128,)	-2.1539	+1.4636	+0.5130
bn3b_branch2b/moving_variance:0	(128,)	+0.0961	+1.6067	+0.2474
res3b_branch2c/kernel:0	(1, 1, 128, 512)	-0.3111	+0.4652	+0.0288
bn3b_branch2c/gamma:0	(512,)	-0.0367	+2.0281	+0.4083
bn3b_branch2c/beta:0	(512,)	-1.5916	+1.1728	+0.3779
bn3b_branch2c/moving_mean:0	(512,)	-1.0827	+0.6886	+0.2301
bn3b_branch2c/moving_variance:0	(512,)	+0.0002	+0.2121	+0.0303
res3c_branch2a/kernel:0	(1, 1, 512, 128)	-0.2663	+0.2643	+0.0284
bn3c_branch2a/gamma:0	(128,)	+0.5964	+1.5047	+0.1916
bn3c_branch2a/beta:0	(128,)	-2.8335	+0.7305	+0.5454
bn3c_branch2a/moving_mean:0	(128,)	-3.0245	+1.7478	+0.8335
bn3c_branch2a/moving_variance:0	(128,)	+0.2705	+3.8971	+0.7585
res3c_branch2b/kernel:0	(3, 3, 128, 128)	-0.2296	+0.1968	+0.0233
bn3c_branch2b/gamma:0	(128,)	+0.4471	+1.5901	+0.2878
bn3c_branch2b/beta:0	(128,)	-1.4295	+1.3293	+0.5597
bn3c_branch2b/moving_mean:0	(128,)	-1.0368	+0.7158	+0.2785
bn3c_branch2b/moving_variance:0	(128,)	+0.1252	+0.9441	+0.1439
res3c_branch2c/kernel:0	(1, 1, 128, 512)	-0.3068	+0.3730	+0.0263
bn3c_branch2c/gamma:0	(512,)	-0.0337	+1.9127	+0.3803
bn3c_branch2c/beta:0	(512,)	-1.5568	+0.8346	+0.3645
bn3c_branch2c/moving_mean:0	(512,)	-0.8692	+0.7091	+0.1843
bn3c_branch2c/moving_variance:0	(512,)	+0.0002	+0.1747	+0.0270
res3d_branch2a/kernel:0	(1, 1, 512, 128)	-0.2611	+0.2868	+0.0307
bn3d_branch2a/gamma:0	(128,)	+0.6513	+1.4620	+0.1939
bn3d_branch2a/beta:0	(128,)	-3.0044	+0.6187	+0.6454
bn3d_branch2a/moving_mean:0	(128,)	-3.7891	+2.2792	+0.9724
bn3d_branch2a/moving_variance:0	(128,)	+0.0035	+3.3270	+0.5634
res3d_branch2b/kernel:0	(3, 3, 128, 128)	-0.1610	+0.2406	+0.0237
bn3d_branch2b/gamma:0	(128,)	+0.6456	+3.2960	+0.2932
bn3d_branch2b/beta:0	(128,)	-1.6672	+1.5725	+0.5648
bn3d_branch2b/moving_mean:0	(128,)	-1.0483	+0.3162	+0.2877
bn3d_branch2b/moving_variance:0	(128,)	+0.2354	+1.6038	+0.2114
res3d_branch2c/kernel:0	(1, 1, 128, 512)	-0.2424	+0.3360	+0.0271
bn3d_branch2c/gamma:0	(512,)	-0.0229	+1.9388	+0.5210
bn3d_branch2c/beta:0	(512,)	-1.0656	+0.9844	+0.2739
bn3d_branch2c/moving_mean:0	(512,)	-1.1224	+0.4736	+0.2424
bn3d_branch2c/moving_variance:0	(512,)	+0.0002	+0.4452	+0.0605
res4a_branch2a/kernel:0	(1, 1, 512, 256)	-0.2850	+0.2532	+0.0148
bn4a_branch2a/gamma:0	(256,)	+0.4807	+1.4740	+0.1509
bn4a_branch2a/beta:0	(256,)	-1.9598	+1.1296	+0.3918
bn4a_branch2a/moving_mean:0	(256,)	-3.6537	+1.2955	+0.6511
bn4a_branch2a/moving_variance:0	(256,)	+0.0614	+2.4605	+0.3051
res4a_branch2b/kernel:0	(3, 3, 256, 256)	-0.1724	+0.1962	+0.0106
bn4a_branch2b/gamma:0	(256,)	+0.4749	+1.5932	+0.2005
bn4a_branch2b/beta:0	(256,)	-2.6223	+1.1253	+0.4850
bn4a_branch2b/moving_mean:0	(256,)	-2.6573	+3.4857	+0.5984
bn4a_branch2b/moving_variance:0	(256,)	+0.1394	+1.4291	+0.2191
res4a_branch2c/kernel:0	(1, 1, 256, 1024)	-0.2826	+0.1895	+0.0140
res4a_branch1/kernel:0	(1, 1, 512, 1024)	-0.3567	+0.3337	+0.0158
bn4a_branch2c/gamma:0	(1024,)	-0.0100	+2.8427	+0.4587
bn4a_branch2c/beta:0	(1024,)	-0.5290	+2.0636	+0.2888
bn4a_branch2c/moving_mean:0	(1024,)	-0.4885	+0.2798	+0.0831
bn4a_branch2c/moving_variance:0	(1024,)	+0.0000	+0.1206	+0.0119
bn4a_branch1/gamma:0	(1024,)	+0.1721	+4.0220	+0.7190
bn4a_branch1/beta:0	(1024,)	-0.5290	+2.0638	+0.2888
bn4a_branch1/moving_mean:0	(1024,)	-5.8239	+3.2843	+0.9039
bn4a_branch1/moving_variance:0	(1024,)	+0.0546	+8.5205	+0.6964
res4b_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1260	+0.1778	+0.0081
bn4b_branch2a/gamma:0	(256,)	+0.4198	+1.6067	+0.1885
bn4b_branch2a/beta:0	(256,)	-2.2254	+2.0613	+0.4898
bn4b_branch2a/moving_mean:0	(256,)	-4.6812	+2.7681	+1.1843
bn4b_branch2a/moving_variance:0	(256,)	+0.1327	+11.9923	+1.3851
res4b_branch2b/kernel:0	(3, 3, 256, 256)	-0.0965	+0.1632	+0.0072
bn4b_branch2b/gamma:0	(256,)	+0.5049	+1.4772	+0.1899
bn4b_branch2b/beta:0	(256,)	-1.6971	+0.5928	+0.4359
bn4b_branch2b/moving_mean:0	(256,)	-5.7095	+1.9171	+0.8444
bn4b_branch2b/moving_variance:0	(256,)	+0.0203	+1.3132	+0.2172
res4b_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1962	+0.2793	+0.0103
bn4b_branch2c/gamma:0	(1024,)	-0.0006	+3.1835	+0.3656
bn4b_branch2c/beta:0	(1024,)	-1.0684	+0.9525	+0.1818
bn4b_branch2c/moving_mean:0	(1024,)	-0.4443	+0.4752	+0.0943
bn4b_branch2c/moving_variance:0	(1024,)	+0.0000	+0.1888	+0.0161
res4c_branch2a/kernel:0	(1, 1, 1024, 256)	-0.0990	+0.1245	+0.0082
bn4c_branch2a/gamma:0	(256,)	+0.5761	+1.7694	+0.1421
bn4c_branch2a/beta:0	(256,)	-0.9332	+1.3302	+0.3766
bn4c_branch2a/moving_mean:0	(256,)	-4.0562	+2.6203	+1.2418
bn4c_branch2a/moving_variance:0	(256,)	+0.3526	+5.5460	+0.6766
res4c_branch2b/kernel:0	(3, 3, 256, 256)	-0.0971	+0.1085	+0.0074
bn4c_branch2b/gamma:0	(256,)	+0.5058	+1.2290	+0.1501
bn4c_branch2b/beta:0	(256,)	-1.4553	+0.5811	+0.3399
bn4c_branch2b/moving_mean:0	(256,)	-3.4174	+1.7898	+0.5966
bn4c_branch2b/moving_variance:0	(256,)	+0.0730	+1.5945	+0.1712
res4c_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1343	+0.1620	+0.0107
bn4c_branch2c/gamma:0	(1024,)	+0.0047	+2.2900	+0.2671
bn4c_branch2c/beta:0	(1024,)	-1.1121	+0.7399	+0.1807
bn4c_branch2c/moving_mean:0	(1024,)	-0.3355	+0.1449	+0.0609
bn4c_branch2c/moving_variance:0	(1024,)	+0.0003	+0.0969	+0.0076
res4d_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1154	+0.1464	+0.0103
bn4d_branch2a/gamma:0	(256,)	+0.5715	+1.4526	+0.1505
bn4d_branch2a/beta:0	(256,)	-1.3573	+0.4802	+0.3064
bn4d_branch2a/moving_mean:0	(256,)	-3.2371	+2.2156	+0.9378
bn4d_branch2a/moving_variance:0	(256,)	+0.3515	+5.9965	+0.7573
res4d_branch2b/kernel:0	(3, 3, 256, 256)	-0.0995	+0.0993	+0.0087
bn4d_branch2b/gamma:0	(256,)	+0.4295	+1.5526	+0.1616
bn4d_branch2b/beta:0	(256,)	-1.4477	+0.3874	+0.2842
bn4d_branch2b/moving_mean:0	(256,)	-1.4094	+0.7436	+0.2812
bn4d_branch2b/moving_variance:0	(256,)	+0.0523	+0.4509	+0.0769
res4d_branch2c/kernel:0	(1, 1, 256, 1024)	-0.2408	+0.1240	+0.0118
bn4d_branch2c/gamma:0	(1024,)	+0.0455	+2.8707	+0.3316
bn4d_branch2c/beta:0	(1024,)	-1.3654	+0.5976	+0.2313
bn4d_branch2c/moving_mean:0	(1024,)	-0.3255	+0.1030	+0.0535
bn4d_branch2c/moving_variance:0	(1024,)	+0.0017	+0.1031	+0.0077
res4e_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1464	+0.1125	+0.0101
bn4e_branch2a/gamma:0	(256,)	+0.6457	+1.3797	+0.1243
bn4e_branch2a/beta:0	(256,)	-1.0967	+0.3581	+0.2712
bn4e_branch2a/moving_mean:0	(256,)	-4.3375	+1.8791	+1.1414
bn4e_branch2a/moving_variance:0	(256,)	+0.3798	+4.6768	+0.6343
res4e_branch2b/kernel:0	(3, 3, 256, 256)	-0.0862	+0.0903	+0.0090
bn4e_branch2b/gamma:0	(256,)	+0.5497	+1.2982	+0.1328
bn4e_branch2b/beta:0	(256,)	-1.2916	+0.2223	+0.2549
bn4e_branch2b/moving_mean:0	(256,)	-1.6195	+0.8106	+0.3150
bn4e_branch2b/moving_variance:0	(256,)	+0.0594	+0.9789	+0.1007
res4e_branch2c/kernel:0	(1, 1, 256, 1024)	-0.2030	+0.1746	+0.0119
bn4e_branch2c/gamma:0	(1024,)	+0.0241	+1.7471	+0.1844
bn4e_branch2c/beta:0	(1024,)	-1.0605	+0.4113	+0.1782
bn4e_branch2c/moving_mean:0	(1024,)	-0.2640	+0.1132	+0.0421
bn4e_branch2c/moving_variance:0	(1024,)	+0.0009	+0.0492	+0.0041
res4f_branch2a/kernel:0	(1, 1, 1024, 256)	-0.0842	+0.1227	+0.0105
bn4f_branch2a/gamma:0	(256,)	+0.7024	+1.3934	+0.1006
bn4f_branch2a/beta:0	(256,)	-1.1660	+0.3933	+0.2485
bn4f_branch2a/moving_mean:0	(256,)	-4.4642	+2.1927	+1.1421
bn4f_branch2a/moving_variance:0	(256,)	+0.4538	+4.7501	+0.7294
res4f_branch2b/kernel:0	(3, 3, 256, 256)	-0.1032	+0.0997	+0.0095
bn4f_branch2b/gamma:0	(256,)	+0.4258	+1.2521	+0.1110
bn4f_branch2b/beta:0	(256,)	-1.4425	+0.5213	+0.2464
bn4f_branch2b/moving_mean:0	(256,)	-1.8044	+1.5505	+0.3778
bn4f_branch2b/moving_variance:0	(256,)	+0.0635	+0.6795	+0.0933
res4f_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1423	+0.1278	+0.0119
bn4f_branch2c/gamma:0	(1024,)	+0.1776	+1.6305	+0.1749
bn4f_branch2c/beta:0	(1024,)	-0.9891	+0.3002	+0.1423
bn4f_branch2c/moving_mean:0	(1024,)	-0.1586	+0.0747	+0.0354
bn4f_branch2c/moving_variance:0	(1024,)	+0.0012	+0.0210	+0.0024
res4g_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1148	+0.2416	+0.0107
bn4g_branch2a/gamma:0	(256,)	+0.6074	+1.2107	+0.1073
bn4g_branch2a/beta:0	(256,)	-1.2790	+0.2364	+0.2842
bn4g_branch2a/moving_mean:0	(256,)	-4.3445	+1.4500	+1.0834
bn4g_branch2a/moving_variance:0	(256,)	+0.3768	+3.8029	+0.7079
res4g_branch2b/kernel:0	(3, 3, 256, 256)	-0.1280	+0.1199	+0.0097
bn4g_branch2b/gamma:0	(256,)	+0.4760	+1.7497	+0.1351
bn4g_branch2b/beta:0	(256,)	-1.2725	+0.1908	+0.2716
bn4g_branch2b/moving_mean:0	(256,)	-1.1725	+1.0331	+0.2961
bn4g_branch2b/moving_variance:0	(256,)	+0.0579	+0.7416	+0.0856
res4g_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1498	+0.2285	+0.0118
bn4g_branch2c/gamma:0	(1024,)	+0.0908	+1.8260	+0.1987
bn4g_branch2c/beta:0	(1024,)	-0.9102	+0.2949	+0.1424
bn4g_branch2c/moving_mean:0	(1024,)	-0.1887	+0.0784	+0.0394
bn4g_branch2c/moving_variance:0	(1024,)	+0.0013	+0.0316	+0.0033
res4h_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1305	+0.1624	+0.0116
bn4h_branch2a/gamma:0	(256,)	+0.6257	+1.2189	+0.0991
bn4h_branch2a/beta:0	(256,)	-1.4250	+0.0732	+0.2633
bn4h_branch2a/moving_mean:0	(256,)	-3.7871	+2.4121	+0.9370
bn4h_branch2a/moving_variance:0	(256,)	+0.5296	+3.3483	+0.5827
res4h_branch2b/kernel:0	(3, 3, 256, 256)	-0.0986	+0.1224	+0.0102
bn4h_branch2b/gamma:0	(256,)	+0.4840	+1.4915	+0.1486
bn4h_branch2b/beta:0	(256,)	-1.5969	+0.4351	+0.2542
bn4h_branch2b/moving_mean:0	(256,)	-1.0446	+1.1061	+0.2066
bn4h_branch2b/moving_variance:0	(256,)	+0.0492	+0.5152	+0.0675
res4h_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1463	+0.2364	+0.0120
bn4h_branch2c/gamma:0	(1024,)	+0.0535	+2.4894	+0.2982
bn4h_branch2c/beta:0	(1024,)	-0.8032	+0.3345	+0.1533
bn4h_branch2c/moving_mean:0	(1024,)	-0.2073	+0.1215	+0.0407
bn4h_branch2c/moving_variance:0	(1024,)	+0.0012	+0.0485	+0.0043
res4i_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1310	+0.2966	+0.0130
bn4i_branch2a/gamma:0	(256,)	+0.3535	+1.0589	+0.1278
bn4i_branch2a/beta:0	(256,)	-1.4845	+0.4568	+0.2795
bn4i_branch2a/moving_mean:0	(256,)	-4.1849	+2.5485	+1.1041
bn4i_branch2a/moving_variance:0	(256,)	+0.7884	+7.4350	+0.6603
res4i_branch2b/kernel:0	(3, 3, 256, 256)	-0.1289	+0.1534	+0.0092
bn4i_branch2b/gamma:0	(256,)	+0.5686	+1.6406	+0.1422
bn4i_branch2b/beta:0	(256,)	-1.4987	+0.4807	+0.2660
bn4i_branch2b/moving_mean:0	(256,)	-0.5127	+0.1199	+0.0935
bn4i_branch2b/moving_variance:0	(256,)	+0.0143	+0.1193	+0.0159
res4i_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1553	+0.1522	+0.0118
bn4i_branch2c/gamma:0	(1024,)	+0.0405	+2.1795	+0.1866
bn4i_branch2c/beta:0	(1024,)	-0.5968	+0.7084	+0.1265
bn4i_branch2c/moving_mean:0	(1024,)	-0.4519	+0.1328	+0.0659
bn4i_branch2c/moving_variance:0	(1024,)	+0.0008	+0.0992	+0.0064
res4j_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1212	+0.1714	+0.0124
bn4j_branch2a/gamma:0	(256,)	+0.5074	+1.2941	+0.1164
bn4j_branch2a/beta:0	(256,)	-1.9232	+0.2171	+0.2896
bn4j_branch2a/moving_mean:0	(256,)	-4.7605	+1.3667	+0.9790
bn4j_branch2a/moving_variance:0	(256,)	+0.8111	+8.3345	+0.7015
res4j_branch2b/kernel:0	(3, 3, 256, 256)	-0.1000	+0.2422	+0.0102
bn4j_branch2b/gamma:0	(256,)	+0.4069	+1.4642	+0.1343
bn4j_branch2b/beta:0	(256,)	-1.9635	+0.4987	+0.3003
bn4j_branch2b/moving_mean:0	(256,)	-1.0420	+0.6202	+0.2047
bn4j_branch2b/moving_variance:0	(256,)	+0.0464	+0.5519	+0.0491
res4j_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1389	+0.1597	+0.0118
bn4j_branch2c/gamma:0	(1024,)	+0.0293	+2.1061	+0.1991
bn4j_branch2c/beta:0	(1024,)	-0.8361	+0.1735	+0.1254
bn4j_branch2c/moving_mean:0	(1024,)	-0.2061	+0.0772	+0.0375
bn4j_branch2c/moving_variance:0	(1024,)	+0.0003	+0.0274	+0.0028
res4k_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1359	+0.1878	+0.0112
bn4k_branch2a/gamma:0	(256,)	+0.5420	+1.2074	+0.1235
bn4k_branch2a/beta:0	(256,)	-1.7435	+0.3985	+0.3122
bn4k_branch2a/moving_mean:0	(256,)	-6.0315	+1.7842	+1.1139
bn4k_branch2a/moving_variance:0	(256,)	+0.3504	+4.6503	+0.6216
res4k_branch2b/kernel:0	(3, 3, 256, 256)	-0.0792	+0.1220	+0.0093
bn4k_branch2b/gamma:0	(256,)	+0.4983	+1.2338	+0.1261
bn4k_branch2b/beta:0	(256,)	-1.2916	+0.1997	+0.2615
bn4k_branch2b/moving_mean:0	(256,)	-1.0697	+1.5169	+0.3087
bn4k_branch2b/moving_variance:0	(256,)	+0.0201	+0.4110	+0.0620
res4k_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1248	+0.2098	+0.0111
bn4k_branch2c/gamma:0	(1024,)	+0.1164	+2.0104	+0.2095
bn4k_branch2c/beta:0	(1024,)	-1.6254	+0.1871	+0.1615
bn4k_branch2c/moving_mean:0	(1024,)	-0.1643	+0.0819	+0.0371
bn4k_branch2c/moving_variance:0	(1024,)	+0.0018	+0.0433	+0.0044
res4l_branch2a/kernel:0	(1, 1, 1024, 256)	-0.2063	+0.1837	+0.0132
bn4l_branch2a/gamma:0	(256,)	+0.4153	+1.5392	+0.1226
bn4l_branch2a/beta:0	(256,)	-1.8618	+0.3083	+0.3002
bn4l_branch2a/moving_mean:0	(256,)	-4.3532	+1.4638	+1.0180
bn4l_branch2a/moving_variance:0	(256,)	+0.6372	+5.7646	+0.7036
res4l_branch2b/kernel:0	(3, 3, 256, 256)	-0.1077	+0.1703	+0.0099
bn4l_branch2b/gamma:0	(256,)	+0.3988	+1.3548	+0.1309
bn4l_branch2b/beta:0	(256,)	-1.4917	+0.4846	+0.2733
bn4l_branch2b/moving_mean:0	(256,)	-0.4929	+0.3995	+0.1252
bn4l_branch2b/moving_variance:0	(256,)	+0.0217	+0.2759	+0.0259
res4l_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1198	+0.1658	+0.0120
bn4l_branch2c/gamma:0	(1024,)	+0.0792	+1.6846	+0.1774
bn4l_branch2c/beta:0	(1024,)	-1.0420	+0.7085	+0.1541
bn4l_branch2c/moving_mean:0	(1024,)	-0.1533	+0.0933	+0.0373
bn4l_branch2c/moving_variance:0	(1024,)	+0.0009	+0.0339	+0.0026
res4m_branch2a/kernel:0	(1, 1, 1024, 256)	-0.0796	+0.1565	+0.0116
bn4m_branch2a/gamma:0	(256,)	+0.5385	+1.1934	+0.0998
bn4m_branch2a/beta:0	(256,)	-1.3115	+0.3087	+0.2261
bn4m_branch2a/moving_mean:0	(256,)	-5.8896	+1.3802	+1.0426
bn4m_branch2a/moving_variance:0	(256,)	+0.5695	+7.6590	+0.7444
res4m_branch2b/kernel:0	(3, 3, 256, 256)	-0.1025	+0.1319	+0.0090
bn4m_branch2b/gamma:0	(256,)	+0.5824	+1.2019	+0.1020
bn4m_branch2b/beta:0	(256,)	-1.3820	+0.1980	+0.2375
bn4m_branch2b/moving_mean:0	(256,)	-0.7877	+0.6563	+0.1876
bn4m_branch2b/moving_variance:0	(256,)	+0.0295	+0.3663	+0.0473
res4m_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1435	+0.1559	+0.0111
bn4m_branch2c/gamma:0	(1024,)	+0.1877	+1.8117	+0.1711
bn4m_branch2c/beta:0	(1024,)	-0.7704	+0.5870	+0.1488
bn4m_branch2c/moving_mean:0	(1024,)	-0.2008	+0.0926	+0.0429
bn4m_branch2c/moving_variance:0	(1024,)	+0.0013	+0.0343	+0.0039
res4n_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1173	+0.1577	+0.0126
bn4n_branch2a/gamma:0	(256,)	+0.4684	+1.0960	+0.1128
bn4n_branch2a/beta:0	(256,)	-1.2844	+0.0426	+0.2344
bn4n_branch2a/moving_mean:0	(256,)	-3.2392	+1.7882	+0.8253
bn4n_branch2a/moving_variance:0	(256,)	+0.6329	+3.4732	+0.4616
res4n_branch2b/kernel:0	(3, 3, 256, 256)	-0.1209	+0.1524	+0.0087
bn4n_branch2b/gamma:0	(256,)	+0.4896	+1.2047	+0.1105
bn4n_branch2b/beta:0	(256,)	-1.0426	+0.6021	+0.2205
bn4n_branch2b/moving_mean:0	(256,)	-0.3883	+0.1123	+0.0881
bn4n_branch2b/moving_variance:0	(256,)	+0.0143	+0.2724	+0.0217
res4n_branch2c/kernel:0	(1, 1, 256, 1024)	-0.0969	+0.1517	+0.0107
bn4n_branch2c/gamma:0	(1024,)	+0.1913	+1.6900	+0.1242
bn4n_branch2c/beta:0	(1024,)	-0.7635	+0.6491	+0.1329
bn4n_branch2c/moving_mean:0	(1024,)	-0.2299	+0.1080	+0.0471
bn4n_branch2c/moving_variance:0	(1024,)	+0.0021	+0.0429	+0.0041
res4o_branch2a/kernel:0	(1, 1, 1024, 256)	-0.0868	+0.1276	+0.0122
bn4o_branch2a/gamma:0	(256,)	+0.4140	+1.0878	+0.1046
bn4o_branch2a/beta:0	(256,)	-1.5212	+0.1588	+0.2378
bn4o_branch2a/moving_mean:0	(256,)	-6.5676	+1.6475	+1.1866
bn4o_branch2a/moving_variance:0	(256,)	+0.6190	+5.9674	+0.6751
res4o_branch2b/kernel:0	(3, 3, 256, 256)	-0.0956	+0.1297	+0.0088
bn4o_branch2b/gamma:0	(256,)	+0.5295	+1.2251	+0.1097
bn4o_branch2b/beta:0	(256,)	-1.2628	+0.4158	+0.2264
bn4o_branch2b/moving_mean:0	(256,)	-0.4420	+0.3259	+0.1114
bn4o_branch2b/moving_variance:0	(256,)	+0.0193	+0.1740	+0.0210
res4o_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1526	+0.1561	+0.0108
bn4o_branch2c/gamma:0	(1024,)	+0.2372	+1.9252	+0.1541
bn4o_branch2c/beta:0	(1024,)	-0.8091	+0.5670	+0.1401
bn4o_branch2c/moving_mean:0	(1024,)	-0.2384	+0.1076	+0.0495
bn4o_branch2c/moving_variance:0	(1024,)	+0.0016	+0.0486	+0.0047
res4p_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1428	+0.1984	+0.0135
bn4p_branch2a/gamma:0	(256,)	+0.5017	+1.0527	+0.0912
bn4p_branch2a/beta:0	(256,)	-1.4882	+0.0455	+0.2393
bn4p_branch2a/moving_mean:0	(256,)	-3.1050	+2.3637	+0.9707
bn4p_branch2a/moving_variance:0	(256,)	+0.7623	+3.6884	+0.5522
res4p_branch2b/kernel:0	(3, 3, 256, 256)	-0.0858	+0.1100	+0.0100
bn4p_branch2b/gamma:0	(256,)	+0.4397	+1.4020	+0.1318
bn4p_branch2b/beta:0	(256,)	-1.4270	+0.4049	+0.2497
bn4p_branch2b/moving_mean:0	(256,)	-0.4054	+0.3653	+0.1037
bn4p_branch2b/moving_variance:0	(256,)	+0.0251	+0.1553	+0.0220
res4p_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1097	+0.1687	+0.0119
bn4p_branch2c/gamma:0	(1024,)	+0.1811	+1.7263	+0.1962
bn4p_branch2c/beta:0	(1024,)	-1.0450	+0.3895	+0.1635
bn4p_branch2c/moving_mean:0	(1024,)	-0.2053	+0.1271	+0.0407
bn4p_branch2c/moving_variance:0	(1024,)	+0.0015	+0.0479	+0.0040
res4q_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1232	+0.2498	+0.0136
bn4q_branch2a/gamma:0	(256,)	+0.3415	+1.0128	+0.1070
bn4q_branch2a/beta:0	(256,)	-1.5989	+0.3609	+0.2903
bn4q_branch2a/moving_mean:0	(256,)	-5.2214	+2.3356	+1.1035
bn4q_branch2a/moving_variance:0	(256,)	+0.6609	+11.6783	+0.9515
res4q_branch2b/kernel:0	(3, 3, 256, 256)	-0.1798	+0.1955	+0.0088
bn4q_branch2b/gamma:0	(256,)	+0.6543	+1.4769	+0.1200
bn4q_branch2b/beta:0	(256,)	-1.1978	+0.3759	+0.2500
bn4q_branch2b/moving_mean:0	(256,)	-0.3519	+0.1123	+0.0780
bn4q_branch2b/moving_variance:0	(256,)	+0.0133	+0.1143	+0.0136
res4q_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1707	+0.1696	+0.0116
bn4q_branch2c/gamma:0	(1024,)	+0.0371	+2.1323	+0.2144
bn4q_branch2c/beta:0	(1024,)	-0.7875	+0.3563	+0.1508
bn4q_branch2c/moving_mean:0	(1024,)	-0.3719	+0.2027	+0.0712
bn4q_branch2c/moving_variance:0	(1024,)	+0.0013	+0.0541	+0.0063
res4r_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1752	+0.2489	+0.0135
bn4r_branch2a/gamma:0	(256,)	+0.2799	+0.9274	+0.1076
bn4r_branch2a/beta:0	(256,)	-1.3579	+0.2758	+0.2674
bn4r_branch2a/moving_mean:0	(256,)	-2.7110	+3.0688	+0.9259
bn4r_branch2a/moving_variance:0	(256,)	+0.8098	+9.7240	+0.7793
res4r_branch2b/kernel:0	(3, 3, 256, 256)	-0.1313	+0.1872	+0.0085
bn4r_branch2b/gamma:0	(256,)	+0.5145	+1.4675	+0.1384
bn4r_branch2b/beta:0	(256,)	-0.9184	+0.6886	+0.2130
bn4r_branch2b/moving_mean:0	(256,)	-0.2939	+0.1038	+0.0677
bn4r_branch2b/moving_variance:0	(256,)	+0.0066	+0.0804	+0.0106
res4r_branch2c/kernel:0	(1, 1, 256, 1024)	-0.0867	+0.1725	+0.0110
bn4r_branch2c/gamma:0	(1024,)	+0.1375	+1.7613	+0.1524
bn4r_branch2c/beta:0	(1024,)	-0.7547	+0.3295	+0.1355
bn4r_branch2c/moving_mean:0	(1024,)	-0.2626	+0.1839	+0.0649
bn4r_branch2c/moving_variance:0	(1024,)	+0.0028	+0.0670	+0.0051
res4s_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1282	+0.1822	+0.0127
bn4s_branch2a/gamma:0	(256,)	+0.3420	+0.9798	+0.1074
bn4s_branch2a/beta:0	(256,)	-1.2975	+0.5379	+0.2706
bn4s_branch2a/moving_mean:0	(256,)	-8.8698	+2.0926	+1.2141
bn4s_branch2a/moving_variance:0	(256,)	+0.6463	+17.2415	+1.1652
res4s_branch2b/kernel:0	(3, 3, 256, 256)	-0.2004	+0.1857	+0.0084
bn4s_branch2b/gamma:0	(256,)	+0.5218	+1.2066	+0.1056
bn4s_branch2b/beta:0	(256,)	-1.1360	+0.3336	+0.2261
bn4s_branch2b/moving_mean:0	(256,)	-0.4850	+0.1296	+0.0932
bn4s_branch2b/moving_variance:0	(256,)	+0.0122	+0.0847	+0.0115
res4s_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1153	+0.1679	+0.0109
bn4s_branch2c/gamma:0	(1024,)	+0.1504	+1.7503	+0.1436
bn4s_branch2c/beta:0	(1024,)	-0.7774	+0.4413	+0.1241
bn4s_branch2c/moving_mean:0	(1024,)	-0.2327	+0.1329	+0.0533
bn4s_branch2c/moving_variance:0	(1024,)	+0.0022	+0.0453	+0.0037
res4t_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1595	+0.1765	+0.0128
bn4t_branch2a/gamma:0	(256,)	+0.4376	+1.2511	+0.1069
bn4t_branch2a/beta:0	(256,)	-1.1352	+0.2663	+0.2456
bn4t_branch2a/moving_mean:0	(256,)	-6.3550	+2.2208	+1.3414
bn4t_branch2a/moving_variance:0	(256,)	+0.7818	+5.8100	+0.6672
res4t_branch2b/kernel:0	(3, 3, 256, 256)	-0.1327	+0.1067	+0.0091
bn4t_branch2b/gamma:0	(256,)	+0.4616	+1.2523	+0.1079
bn4t_branch2b/beta:0	(256,)	-1.1122	+0.6912	+0.2212
bn4t_branch2b/moving_mean:0	(256,)	-0.8846	+0.4358	+0.1821
bn4t_branch2b/moving_variance:0	(256,)	+0.0364	+0.3580	+0.0398
res4t_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1665	+0.1575	+0.0114
bn4t_branch2c/gamma:0	(1024,)	+0.2199	+2.0075	+0.1730
bn4t_branch2c/beta:0	(1024,)	-0.7963	+0.3039	+0.1346
bn4t_branch2c/moving_mean:0	(1024,)	-0.2121	+0.1756	+0.0493
bn4t_branch2c/moving_variance:0	(1024,)	+0.0014	+0.0391	+0.0036
res4u_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1065	+0.1518	+0.0119
bn4u_branch2a/gamma:0	(256,)	+0.2817	+1.1194	+0.1175
bn4u_branch2a/beta:0	(256,)	-1.3112	+0.5442	+0.2535
bn4u_branch2a/moving_mean:0	(256,)	-9.9501	+2.8899	+1.3347
bn4u_branch2a/moving_variance:0	(256,)	+0.4432	+12.1801	+1.2243
res4u_branch2b/kernel:0	(3, 3, 256, 256)	-0.1258	+0.1178	+0.0078
bn4u_branch2b/gamma:0	(256,)	+0.6214	+1.3642	+0.1069
bn4u_branch2b/beta:0	(256,)	-0.9785	+0.4391	+0.2048
bn4u_branch2b/moving_mean:0	(256,)	-0.5389	+0.4149	+0.1213
bn4u_branch2b/moving_variance:0	(256,)	+0.0170	+0.1117	+0.0160
res4u_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1381	+0.1941	+0.0103
bn4u_branch2c/gamma:0	(1024,)	+0.0992	+1.7961	+0.1407
bn4u_branch2c/beta:0	(1024,)	-0.7827	+0.7017	+0.1639
bn4u_branch2c/moving_mean:0	(1024,)	-0.2808	+0.1409	+0.0702
bn4u_branch2c/moving_variance:0	(1024,)	+0.0023	+0.0837	+0.0066
res4v_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1570	+0.2220	+0.0123
bn4v_branch2a/gamma:0	(256,)	+0.3942	+1.0195	+0.0944
bn4v_branch2a/beta:0	(256,)	-1.2374	+0.4526	+0.2716
bn4v_branch2a/moving_mean:0	(256,)	-6.7398	+2.1281	+1.2705
bn4v_branch2a/moving_variance:0	(256,)	+0.6142	+6.2192	+0.7720
res4v_branch2b/kernel:0	(3, 3, 256, 256)	-0.1412	+0.1655	+0.0083
bn4v_branch2b/gamma:0	(256,)	+0.6196	+1.1648	+0.0919
bn4v_branch2b/beta:0	(256,)	-1.0248	+0.8823	+0.1907
bn4v_branch2b/moving_mean:0	(256,)	-0.5943	+0.2525	+0.1225
bn4v_branch2b/moving_variance:0	(256,)	+0.0272	+0.2850	+0.0257
res4v_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1068	+0.1662	+0.0109
bn4v_branch2c/gamma:0	(1024,)	+0.2345	+1.6791	+0.1361
bn4v_branch2c/beta:0	(1024,)	-0.9361	+0.5179	+0.2008
bn4v_branch2c/moving_mean:0	(1024,)	-0.3138	+0.2207	+0.0648
bn4v_branch2c/moving_variance:0	(1024,)	+0.0023	+0.0550	+0.0047
res4w_branch2a/kernel:0	(1, 1, 1024, 256)	-0.1353	+0.2094	+0.0127
bn4w_branch2a/gamma:0	(256,)	+0.2443	+1.0933	+0.1136
bn4w_branch2a/beta:0	(256,)	-1.4777	+0.3498	+0.2970
bn4w_branch2a/moving_mean:0	(256,)	-15.1299	+2.8008	+2.1165
bn4w_branch2a/moving_variance:0	(256,)	+0.7279	+19.8808	+1.5136
res4w_branch2b/kernel:0	(3, 3, 256, 256)	-0.1029	+0.1717	+0.0083
bn4w_branch2b/gamma:0	(256,)	+0.7102	+1.4165	+0.0997
bn4w_branch2b/beta:0	(256,)	-0.9754	+0.3929	+0.2028
bn4w_branch2b/moving_mean:0	(256,)	-0.3679	+0.2234	+0.0824
bn4w_branch2b/moving_variance:0	(256,)	+0.0133	+0.1334	+0.0153
res4w_branch2c/kernel:0	(1, 1, 256, 1024)	-0.1451	+0.1874	+0.0109
bn4w_branch2c/gamma:0	(1024,)	+0.0215	+1.5528	+0.1530
bn4w_branch2c/beta:0	(1024,)	-0.8591	+0.5082	+0.1832
bn4w_branch2c/moving_mean:0	(1024,)	-0.4728	+0.1823	+0.1269
bn4w_branch2c/moving_variance:0	(1024,)	+0.0008	+0.1209	+0.0090
res5a_branch2a/kernel:0	(1, 1, 1024, 512)	-0.1747	+0.2130	+0.0143
bn5a_branch2a/gamma:0	(512,)	+0.5045	+1.2405	+0.1245
bn5a_branch2a/beta:0	(512,)	-1.4638	+0.5064	+0.3062
bn5a_branch2a/moving_mean:0	(512,)	-11.4545	+4.6925	+1.6878
bn5a_branch2a/moving_variance:0	(512,)	+1.1314	+15.7920	+1.4502
res5a_branch2b/kernel:0	(3, 3, 512, 512)	-0.2464	+0.3249	+0.0091
bn5a_branch2b/gamma:0	(512,)	+0.2982	+1.4230	+0.1373
bn5a_branch2b/beta:0	(512,)	-1.6618	+0.7232	+0.3175
bn5a_branch2b/moving_mean:0	(512,)	-2.2210	+1.7648	+0.3094
bn5a_branch2b/moving_variance:0	(512,)	+0.1187	+1.4766	+0.1865
res5a_branch2c/kernel:0	(1, 1, 512, 2048)	-0.2871	+0.3263	+0.0122
res5a_branch1/kernel:0	(1, 1, 1024, 2048)	-0.3741	+0.4705	+0.0105
bn5a_branch2c/gamma:0	(2048,)	+0.6692	+2.7116	+0.2395
bn5a_branch2c/beta:0	(2048,)	-1.8662	+1.4781	+0.2376
bn5a_branch2c/moving_mean:0	(2048,)	-0.5774	+0.6525	+0.0644
bn5a_branch2c/moving_variance:0	(2048,)	+0.0024	+0.1612	+0.0084
bn5a_branch1/gamma:0	(2048,)	+0.8662	+4.8957	+0.5145
bn5a_branch1/beta:0	(2048,)	-1.8661	+1.4784	+0.2376
bn5a_branch1/moving_mean:0	(2048,)	-10.0727	+4.4287	+1.0954
bn5a_branch1/moving_variance:0	(2048,)	+0.2868	+7.7099	+0.5971
res5b_branch2a/kernel:0	(1, 1, 2048, 512)	-0.1615	+0.2535	+0.0106
bn5b_branch2a/gamma:0	(512,)	+0.3789	+1.1436	+0.0969
bn5b_branch2a/beta:0	(512,)	-1.1929	+0.6042	+0.1990
bn5b_branch2a/moving_mean:0	(512,)	-4.4332	+5.7965	+0.6776
bn5b_branch2a/moving_variance:0	(512,)	+0.8363	+7.6290	+0.8946
res5b_branch2b/kernel:0	(3, 3, 512, 512)	-0.1333	+0.2426	+0.0079
bn5b_branch2b/gamma:0	(512,)	+0.5274	+1.1794	+0.1060
bn5b_branch2b/beta:0	(512,)	-1.8549	+0.5551	+0.2810
bn5b_branch2b/moving_mean:0	(512,)	-1.3069	+1.7223	+0.2265
bn5b_branch2b/moving_variance:0	(512,)	+0.0572	+1.0433	+0.0724
res5b_branch2c/kernel:0	(1, 1, 512, 2048)	-0.1352	+0.1977	+0.0106
bn5b_branch2c/gamma:0	(2048,)	+0.5679	+2.4291	+0.2254
bn5b_branch2c/beta:0	(2048,)	-2.3769	+0.1646	+0.2141
bn5b_branch2c/moving_mean:0	(2048,)	-0.4428	+1.0842	+0.0525
bn5b_branch2c/moving_variance:0	(2048,)	+0.0020	+0.2272	+0.0065
res5c_branch2a/kernel:0	(1, 1, 2048, 512)	-0.1994	+0.3588	+0.0115
bn5c_branch2a/gamma:0	(512,)	+0.1406	+1.1445	+0.0988
bn5c_branch2a/beta:0	(512,)	-1.4070	+0.8456	+0.2542
bn5c_branch2a/moving_mean:0	(512,)	-3.1862	+5.4617	+0.5412
bn5c_branch2a/moving_variance:0	(512,)	+0.5824	+9.3422	+1.1140
res5c_branch2b/kernel:0	(3, 3, 512, 512)	-0.0942	+0.0989	+0.0071
bn5c_branch2b/gamma:0	(512,)	+0.4871	+1.1538	+0.0927
bn5c_branch2b/beta:0	(512,)	-1.4373	+0.3356	+0.2770
bn5c_branch2b/moving_mean:0	(512,)	-0.6456	+0.1676	+0.1068
bn5c_branch2b/moving_variance:0	(512,)	+0.0354	+0.4077	+0.0522
res5c_branch2c/kernel:0	(1, 1, 512, 2048)	-0.1317	+0.1323	+0.0103
bn5c_branch2c/gamma:0	(2048,)	+0.6058	+2.5600	+0.2211
bn5c_branch2c/beta:0	(2048,)	-4.0471	-0.6726	+0.2231
bn5c_branch2c/moving_mean:0	(2048,)	-0.3058	+0.1791	+0.0368
bn5c_branch2c/moving_variance:0	(2048,)	+0.0024	+0.0645	+0.0039
fpn_c5p5/kernel:0	(1, 1, 2048, 256)	-0.0507	+0.0571	+0.0073
fpn_c5p5/bias:0	(256,)	-0.0140	+0.0120	+0.0050
fpn_c4p4/kernel:0	(1, 1, 1024, 256)	-0.1139	+0.0834	+0.0094
fpn_c4p4/bias:0	(256,)	-0.0045	+0.0039	+0.0014
fpn_c3p3/kernel:0	(1, 1, 512, 256)	-0.0509	+0.0533	+0.0073
fpn_c3p3/bias:0	(256,)	-0.0061	+0.0055	+0.0020
fpn_c2p2/kernel:0	(1, 1, 256, 256)	-0.0333	+0.0489	+0.0057
fpn_c2p2/bias:0	(256,)	-0.0051	+0.0063	+0.0020
fpn_p5/kernel:0	(3, 3, 256, 256)	-0.0338	+0.0384	+0.0054
fpn_p5/bias:0	(256,)	-0.0080	+0.0079	+0.0035
fpn_p2/kernel:0	(3, 3, 256, 256)	-0.0278	+0.0344	+0.0051
fpn_p2/bias:0	(256,)	-0.0068	+0.0058	+0.0022
fpn_p3/kernel:0	(3, 3, 256, 256)	-0.0246	+0.0288	+0.0046
fpn_p3/bias:0	(256,)	-0.0039	+0.0038	+0.0015
fpn_p4/kernel:0	(3, 3, 256, 256)	-0.0277	+0.0321	+0.0049
fpn_p4/bias:0	(256,)	-0.0038	+0.0034	+0.0016
rpn_conv_shared/kernel:0	(3, 3, 256, 512)	-0.0162	+0.0160	+0.0011
rpn_conv_shared/bias:0	(512,)	-0.0012	+0.0030	+0.0004
rpn_class_raw/kernel:0	(1, 1, 512, 6)	-0.0700	+0.0700	+0.0094
rpn_class_raw/bias:0	(6,)	-0.0092	+0.0092	+0.0054
rpn_bbox_pred/kernel:0	(1, 1, 512, 12)	-0.0844	+0.1395	+0.0139
rpn_bbox_pred/bias:0	(12,)	-0.0167	+0.0201	+0.0097
mrcnn_class_conv1/kernel:0	(7, 7, 256, 1024)	-0.0240	+0.0250	+0.0032
mrcnn_class_conv1/bias:0	(1024,)	-0.0011	+0.0003	+0.0002
mrcnn_class_bn1/gamma:0	(1024,)	+0.9650	+1.0500	+0.0079
mrcnn_class_bn1/beta:0	(1024,)	-0.0320	+0.0050	+0.0033
mrcnn_class_bn1/moving_mean:0	(1024,)	-20.4790	+8.7007	+2.2488
mrcnn_class_bn1/moving_variance:0	(1024,)	+3.2710	+175.0298	+10.5956
mrcnn_class_conv2/kernel:0	(1, 1, 1024, 1024)	-0.0632	+0.0426	+0.0051
mrcnn_class_conv2/bias:0	(1024,)	-0.0156	+0.0214	+0.0039
mrcnn_class_bn2/gamma:0	(1024,)	+0.9801	+1.0526	+0.0089
mrcnn_class_bn2/beta:0	(1024,)	-0.0126	+0.0288	+0.0043
mrcnn_class_bn2/moving_mean:0	(1024,)	-0.6272	+0.5237	+0.1323
mrcnn_class_bn2/moving_variance:0	(1024,)	+0.0072	+0.6561	+0.0431
mrcnn_class_logits/kernel:0	(1024, 4)	-0.0815	+0.0813	+0.0437
mrcnn_class_logits/bias:0	(4,)	-0.0006	+0.0008	+0.0005
mrcnn_bbox_fc/kernel:0	(1024, 16)	-0.0790	+0.0765	+0.0440
mrcnn_bbox_fc/bias:0	(16,)	-0.0010	+0.0007	+0.0004
mrcnn_mask_conv1/kernel:0	(3, 3, 256, 256)	-0.0520	+0.0462	+0.0045
mrcnn_mask_conv1/bias:0	(256,)	-0.0037	+0.0022	+0.0009
mrcnn_mask_bn1/gamma:0	(256,)	+0.9804	+1.0889	+0.0122
mrcnn_mask_bn1/beta:0	(256,)	-0.0216	+0.0028	+0.0036
mrcnn_mask_bn1/moving_mean:0	(256,)	-4.2402	+1.4667	+0.7274
mrcnn_mask_bn1/moving_variance:0	(256,)	+0.3057	+5.8205	+0.9377
mrcnn_mask_conv2/kernel:0	(3, 3, 256, 256)	-0.0544	+0.0949	+0.0045
mrcnn_mask_conv2/bias:0	(256,)	-0.0049	+0.0035	+0.0016
mrcnn_mask_bn2/gamma:0	(256,)	+0.9846	+1.0401	+0.0092
mrcnn_mask_bn2/beta:0	(256,)	-0.0176	+0.0024	+0.0034
mrcnn_mask_bn2/moving_mean:0	(256,)	-0.7085	+0.2866	+0.1437
mrcnn_mask_bn2/moving_variance:0	(256,)	+0.0300	+0.3525	+0.0369
mrcnn_mask_conv3/kernel:0	(3, 3, 256, 256)	-0.0416	+0.0459	+0.0042
mrcnn_mask_conv3/bias:0	(256,)	-0.0107	+0.0073	+0.0028
mrcnn_mask_bn3/gamma:0	(256,)	+0.9867	+1.0359	+0.0074
mrcnn_mask_bn3/beta:0	(256,)	-0.0312	+0.0009	+0.0044
mrcnn_mask_bn3/moving_mean:0	(256,)	-0.5781	+0.2730	+0.1416
mrcnn_mask_bn3/moving_variance:0	(256,)	+0.0265	+0.1663	+0.0258
mrcnn_mask_conv4/kernel:0	(3, 3, 256, 256)	-0.0326	+0.0267	+0.0037
mrcnn_mask_conv4/bias:0	(256,)	-0.0014	+0.0049	+0.0009
mrcnn_mask_bn4/gamma:0	(256,)	+0.9985	+1.0724	+0.0184
mrcnn_mask_bn4/beta:0	(256,)	+0.0042	+0.0456	+0.0111
mrcnn_mask_bn4/moving_mean:0	(256,)	-0.2408	+0.1736	+0.0773
mrcnn_mask_bn4/moving_variance:0	(256,)	+0.0150	+0.0639	+0.0084
mrcnn_mask_deconv/kernel:0	(2, 2, 256, 256)	-0.0273	+0.0518	+0.0047
mrcnn_mask_deconv/bias:0	(256,)	-0.0029	+0.0718	+0.0104
mrcnn_mask/kernel:0	(1, 1, 256, 4)	-0.1607	+0.1528	+0.0875
mrcnn_mask/bias:0	(4,)	-0.0078	+0.0000	+0.0033

Histograms of Weights

TODO: cleanup this part

# Pick layer types to display
LAYER_TYPES = ['Conv2D', 'Dense', 'Conv2DTranspose']
# Get layers
layers = model.get_trainable_layers()
layers = list(filter(lambda l: l.__class__.__name__ in LAYER_TYPES, 
                layers))
# Display Histograms
fig, ax = plt.subplots(len(layers), 2, figsize=(10, 3*len(layers)),
                       gridspec_kw={"hspace":1})
for l, layer in enumerate(layers):
    weights = layer.get_weights()
    for w, weight in enumerate(weights):
        tensor = layer.weights[w]
        ax[l, w].set_title(tensor.name)
        _ = ax[l, w].hist(weight[w].flatten(), 50)