CAR DAMAGE DETECTION USING MASK-RCNN
Abstract
Car damage reporting and penalty calculations have always been a challenging issue for license companies and used car selling companies. This paper deals with the issue of the quantitative analysis of the damages by performing unbiased pricing by using Mask RCNN, the state of the art technology for instance segmentation. This paper is an extension of the business technologies to detect and quantify car scratches to address the problems faced by the used car industry and car rental companies. It will support businesses eliminating middle-men and paving the way for a more objective system of pricing and insurance in the vehicle dealership market. Introduction Instance segmentation is the task of detecting and delineating each distinct object of interest appearing in an image. Mask-RCNN is the current state of the art technology for highly accurate mask detection for RoIs (Region of Interest) .In this project we train the MRCNN model to train and detect effective damage area in an image/video . This project is a business extension of existing technologies to detect car scratches and quantifying damages, in order to tackle the problems faced by used car industry and car rental companies for automation of penalty occurred due to these accidents.
This jupyter notebook contains data visualization of car damage images and automated car damage detection example. First we need to import all the packages including custom functions of Matterport Mask R-CNN’ repository
import os
import sys
import itertools
import math
import logging
import json
import re
import random
from collections import OrderedDict
import numpy as np
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.patches as patches
import matplotlib.lines as lines
from matplotlib.patches import Polygon
# Import Mask RCNN
#sys.path.append(ROOT_DIR) # To find local version of the library
from mrcnn import utils
from mrcnn import visualize
from mrcnn.visualize import display_images
from mrcnn import model
import mrcnn.model as modellib
from mrcnn.model import log
import cv2
import custom,custom_1
import imgaug,h5py,IPython
%matplotlib inline
C:\Users\Sourish\Anaconda3\lib\site-packages\dask\config.py:168: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details.
data = yaml.load(f.read()) or {}
Using TensorFlow backend.
Setting up the configuration
root directory,data path setting up the ,log file path and model object(weight matrix)for inference (prediction)
# Root directory of the project
ROOT_DIR = os.getcwd()
sys.path.append(ROOT_DIR) # To find local version of the library
MODEL_DIR = os.path.join(ROOT_DIR, "logs")
custom_WEIGHTS_PATH = "mask_rcnn_scratch_0013.h5" # TODO: update this path for best performing iteration weights
config = custom.CustomConfig()
custom_DIR = os.path.join(ROOT_DIR, "custom/")
custom_DIR
'C:\\Users\\Sourish\\Mask_RCNN\\custom/'
Loading the Data
# Load dataset
dataset = custom_1.CustomDataset()
dataset.load_custom(custom_DIR, "train")
# Must call before using the dataset
dataset.prepare()
print("Image Count: {}".format(len(dataset.image_ids)))
print("Class Count: {}".format(dataset.num_classes))
for i, info in enumerate(dataset.class_info):
print("{:3}. {:50}".format(i, info['name']))
Image Count: 49
Class Count: 2
0. BG
1. scratch
We will visualize few car damage (scratch) images
# Load and display random samples
image_ids = np.random.choice(dataset.image_ids, 5)
for image_id in image_ids:
image = dataset.load_image(image_id)
mask, class_ids = dataset.load_mask(image_id)
visualize.display_top_masks(image, mask, class_ids, dataset.class_names)
Next we will see Bounding Box(BB)with annotated damage mask for a typical car image.
image_id = random.choice(dataset.image_ids)
image = dataset.load_image(image_id)
mask, class_ids = dataset.load_mask(image_id)
# Compute Bounding box
bbox = utils.extract_bboxes(mask)
# Display image and additional stats
print("image_id ", image_id, dataset.image_reference(image_id))
log("image", image)
log("mask", mask)
log("class_ids", class_ids)
log("bbox", bbox)
# Display image and instances
visualize.display_instances(image, bbox, mask, class_ids, dataset.class_names)
image_id 32 C:\Users\Sourish\Mask_RCNN\custom/train\image35.jpg
image shape: (224, 225, 3) min: 0.00000 max: 255.00000 uint8
mask shape: (224, 225, 2) min: 0.00000 max: 1.00000 bool
class_ids shape: (2,) min: 1.00000 max: 1.00000 int32
bbox shape: (2, 4) min: 81.00000 max: 199.00000 int32
We see some the components of image annotations. Mainly it has x and y co-ordinate of all labeled damages(‘polygon’) and class name(here ‘scratch’) for respective car image.
#Annotation file load
annotations1 = json.load(open(os.path.join(ROOT_DIR, "via_region_data.json"),encoding="utf8"))
annotations = list(annotations1.values())
annotations = [a for a in annotations if a['regions']]
annotations[0]
{'fileref': '',
'size': 46041,
'filename': 'image2.jpg',
'base64_img_data': '',
'file_attributes': {},
'regions': {'0': {'shape_attributes': {'name': 'polygon',
'all_points_x': [428,
429,
480,
518,
557,
577,
610,
660,
642,
578,
579,
585,
590,
574,
580,
516,
507,
474,
427,
426,
412,
412,
430,
470,
452,
428],
'all_points_y': [232,
216,
198,
193,
212,
238,
237,
242,
248,
248,
260,
292,
343,
409,
417,
441,
443,
427,
413,
381,
324,
301,
288,
249,
231,
232]},
'region_attributes': {'Scratch': 'scratch'}},
'1': {'shape_attributes': {'name': 'polygon',
'all_points_x': [470, 500, 578, 718, 670, 594, 553, 510, 469, 448, 470],
'all_points_y': [516, 548, 562, 557, 569, 595, 587, 600, 576, 552, 516]},
'region_attributes': {'Scratch': 'scratch'}}}} #### If we have to quantify a car damage,we need to know the x and y coordinates of the polygon to calculate area of the marked/detected damage.This is for 2nd damage polygon of 'image2.jpg' ```python annotations[1]['regions']['0']['shape_attributes'] l = [] for d in annotations[1]['regions']['0']['shape_attributes'].values():
l.append(d) display('x co-ordinates of the damage:',l[1]) display('y co-ordinates of the damage:',l[2]) ```
'x co-ordinates of the damage:'
[293, 360, 349, 308, 293]
'y co-ordinates of the damage:'
[303, 330, 314, 302, 303] **For prediction or damage detection we need to use the model as inference mode. Model description is consists of important model information like CNN architecture name('resnet101'), ROI threshold(0.9 as defined),configuration description, weightage of different loss components, mask shape, WEIGHT_DECAY etc.**
Get Inferences
config = custom.CustomConfig()
ROOT_DIR = 'C:/Users/Sourish/Mask_RCNN'
CUSTOM_DIR = os.path.join(ROOT_DIR + "/custom/")
print(CUSTOM_DIR)
class InferenceConfig(config.__class__):
# Run detection on one image at a time
GPU_COUNT = 1
IMAGES_PER_GPU = 1
config = InferenceConfig()
config.display()
# 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
# Inspect the model in training or inference modes
# values: 'inference' or 'training'
# TODO: code for 'training' test mode not ready yet
TEST_MODE = "inference"
C:/Users/Sourish/Mask_RCNN/custom/
Configurations:
BACKBONE resnet101
BACKBONE_STRIDES [4, 8, 16, 32, 64]
BATCH_SIZE 1
BBOX_STD_DEV [0.1 0.1 0.2 0.2]
COMPUTE_BACKBONE_SHAPE None
DETECTION_MAX_INSTANCES 100
DETECTION_MIN_CONFIDENCE 0.9
DETECTION_NMS_THRESHOLD 0.3
FPN_CLASSIF_FC_LAYERS_SIZE 1024
GPU_COUNT 1
GRADIENT_CLIP_NORM 5.0
IMAGES_PER_GPU 1
IMAGE_CHANNEL_COUNT 3
IMAGE_MAX_DIM 1024
IMAGE_META_SIZE 14
IMAGE_MIN_DIM 800
IMAGE_MIN_SCALE 0
IMAGE_RESIZE_MODE square
IMAGE_SHAPE [1024 1024 3]
LEARNING_MOMENTUM 0.9
LEARNING_RATE 0.001
LOSS_WEIGHTS {'rpn_class_loss': 1.0, 'rpn_bbox_loss': 1.0, 'mrcnn_class_loss': 1.0, 'mrcnn_bbox_loss': 1.0, 'mrcnn_mask_loss': 1.0}
MASK_POOL_SIZE 14
MASK_SHAPE [28, 28]
MAX_GT_INSTANCES 100
MEAN_PIXEL [123.7 116.8 103.9]
MINI_MASK_SHAPE (56, 56)
NAME damage
NUM_CLASSES 2
POOL_SIZE 7
POST_NMS_ROIS_INFERENCE 1000
POST_NMS_ROIS_TRAINING 2000
PRE_NMS_LIMIT 6000
ROI_POSITIVE_RATIO 0.33
RPN_ANCHOR_RATIOS [0.5, 1, 2]
RPN_ANCHOR_SCALES (32, 64, 128, 256, 512)
RPN_ANCHOR_STRIDE 1
RPN_BBOX_STD_DEV [0.1 0.1 0.2 0.2]
RPN_NMS_THRESHOLD 0.7
RPN_TRAIN_ANCHORS_PER_IMAGE 256
STEPS_PER_EPOCH 100
TOP_DOWN_PYRAMID_SIZE 256
TRAIN_BN False
TRAIN_ROIS_PER_IMAGE 200
USE_MINI_MASK True
USE_RPN_ROIS True
VALIDATION_STEPS 50
WEIGHT_DECAY 0.0001
Helper functions
To visualize predicted damage masks and loading the model weights for prediction
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
from importlib import reload # was constantly changin the visualization, so I decided to reload it instead of notebook
reload(visualize)
# Create model in inference mode
import tensorflow as tf
with tf.device(DEVICE):
model = modellib.MaskRCNN(mode="inference", model_dir=MODEL_DIR,
config=config)
# load the last best model you trained
# weights_path = model.find_last()[1]
custom_WEIGHTS_PATH = 'C:/Users/Sourish/Mask_RCNN/logs/scratch20190612T2046/mask_rcnn_scratch_0013.h5'
# Load weights
print("Loading weights ", custom_WEIGHTS_PATH)
model.load_weights(custom_WEIGHTS_PATH, by_name=True)
Loading weights C:/Users/Sourish/Mask_RCNN/logs/scratch20190612T2046/mask_rcnn_scratch_0013.h5
Re-starting from epoch 13
Loading validation data-set for prediction
dataset = custom_1.CustomDataset()
dataset.load_custom(CUSTOM_DIR,'val')
dataset.prepare()
print('Images: {}\nclasses: {}'.format(len(dataset.image_ids), dataset.class_names))
Images: 6
classes: ['BG', 'scratch']
Visualize model weight
matrix descriptive statistics(shapes, histograms)
visualize.display_weight_stats(model)
WEIGHT NAME | SHAPE | MIN | MAX | STD |
conv1/kernel:0 | (7, 7, 3, 64) | -0.8616 | +0.8451 | +0.1315 |
conv1/bias:0 | (64,) | -0.0002 | +0.0004 | +0.0001 |
bn_conv1/gamma:0 | (64,) | +0.0835 | +2.6411 | +0.5091 |
bn_conv1/beta:0 | (64,) | -2.3931 | +5.3610 | +1.9781 |
bn_conv1/moving_mean:0 | (64,) | -173.0470 | +116.3013 | +44.5654 |
bn_conv1/moving_variance:0*** Overflow? | (64,) | +0.0000 | +146335.3594 | +21847.9668 |
res2a_branch2a/kernel:0 | (1, 1, 64, 64) | -0.6574 | +0.3179 | +0.0764 |
res2a_branch2a/bias:0 | (64,) | -0.0022 | +0.0082 | +0.0018 |
bn2a_branch2a/gamma:0 | (64,) | +0.2169 | +1.8489 | +0.4116 |
bn2a_branch2a/beta:0 | (64,) | -2.1180 | +3.7332 | +1.1786 |
bn2a_branch2a/moving_mean:0 | (64,) | -6.1235 | +7.2220 | +2.2789 |
bn2a_branch2a/moving_variance:0 | (64,) | +0.0000 | +8.9258 | +2.0314 |
res2a_branch2b/kernel:0 | (3, 3, 64, 64) | -0.3878 | +0.5070 | +0.0323 |
res2a_branch2b/bias:0 | (64,) | -0.0037 | +0.0026 | +0.0010 |
bn2a_branch2b/gamma:0 | (64,) | +0.3165 | +1.7010 | +0.3042 |
bn2a_branch2b/beta:0 | (64,) | -1.9348 | +4.5429 | +1.5113 |
bn2a_branch2b/moving_mean:0 | (64,) | -6.7752 | +4.5769 | +2.2594 |
bn2a_branch2b/moving_variance:0 | (64,) | +0.0000 | +5.5085 | +1.0835 |
res2a_branch2c/kernel:0 | (1, 1, 64, 256) | -0.4468 | +0.3615 | +0.0410 |
res2a_branch2c/bias:0 | (256,) | -0.0041 | +0.0052 | +0.0016 |
res2a_branch1/kernel:0 | (1, 1, 64, 256) | -0.8674 | +0.7588 | +0.0703 |
res2a_branch1/bias:0 | (256,) | -0.0034 | +0.0025 | +0.0009 |
bn2a_branch2c/gamma:0 | (256,) | -0.5782 | +3.1806 | +0.6192 |
bn2a_branch2c/beta:0 | (256,) | -1.1422 | +1.4273 | +0.4229 |
bn2a_branch2c/moving_mean:0 | (256,) | -4.2602 | +3.0864 | +1.0168 |
bn2a_branch2c/moving_variance:0 | (256,) | +0.0000 | +2.6688 | +0.3827 |
bn2a_branch1/gamma:0 | (256,) | +0.2411 | +3.4973 | +0.6241 |
bn2a_branch1/beta:0 | (256,) | -1.1422 | +1.4274 | +0.4229 |
bn2a_branch1/moving_mean:0 | (256,) | -8.0883 | +8.6554 | +2.0289 |
bn2a_branch1/moving_variance:0 | (256,) | +0.0000 | +8.7306 | +1.5526 |
res2b_branch2a/kernel:0 | (1, 1, 256, 64) | -0.2536 | +0.2319 | +0.0358 |
res2b_branch2a/bias:0 | (64,) | -0.0027 | +0.0028 | +0.0012 |
bn2b_branch2a/gamma:0 | (64,) | +0.2032 | +1.7708 | +0.3812 |
bn2b_branch2a/beta:0 | (64,) | -2.0546 | +1.6670 | +0.8851 |
bn2b_branch2a/moving_mean:0 | (64,) | -1.5484 | +1.7334 | +0.7177 |
bn2b_branch2a/moving_variance:0 | (64,) | +0.0000 | +2.7921 | +0.7575 |
res2b_branch2b/kernel:0 | (3, 3, 64, 64) | -0.5226 | +0.3397 | +0.0356 |
res2b_branch2b/bias:0 | (64,) | -0.0047 | +0.0033 | +0.0015 |
bn2b_branch2b/gamma:0 | (64,) | +0.5213 | +1.4725 | +0.2252 |
bn2b_branch2b/beta:0 | (64,) | -2.4533 | +2.7526 | +1.1960 |
bn2b_branch2b/moving_mean:0 | (64,) | -1.8186 | +0.8886 | +0.5529 |
bn2b_branch2b/moving_variance:0 | (64,) | +0.0808 | +1.1064 | +0.2187 |
res2b_branch2c/kernel:0 | (1, 1, 64, 256) | -0.3382 | +0.3298 | +0.0415 |
res2b_branch2c/bias:0 | (256,) | -0.0075 | +0.0103 | +0.0020 |
bn2b_branch2c/gamma:0 | (256,) | -0.0363 | +1.7920 | +0.4227 |
bn2b_branch2c/beta:0 | (256,) | -1.2938 | +0.9636 | +0.3430 |
bn2b_branch2c/moving_mean:0 | (256,) | -2.4192 | +2.0440 | +0.5019 |
bn2b_branch2c/moving_variance:0 | (256,) | +0.0000 | +0.1844 | +0.0315 |
res2c_branch2a/kernel:0 | (1, 1, 256, 64) | -0.3012 | +0.2199 | +0.0415 |
res2c_branch2a/bias:0 | (64,) | -0.0009 | +0.0024 | +0.0008 |
bn2c_branch2a/gamma:0 | (64,) | +0.2659 | +1.8204 | +0.2834 |
bn2c_branch2a/beta:0 | (64,) | -2.0168 | +0.8445 | +0.7879 |
bn2c_branch2a/moving_mean:0 | (64,) | -4.5208 | +1.6091 | +1.2391 |
bn2c_branch2a/moving_variance:0 | (64,) | +0.0000 | +3.4581 | +0.7942 |
res2c_branch2b/kernel:0 | (3, 3, 64, 64) | -0.2007 | +0.2176 | +0.0378 |
res2c_branch2b/bias:0 | (64,) | -0.0030 | +0.0058 | +0.0018 |
bn2c_branch2b/gamma:0 | (64,) | +0.6267 | +1.5415 | +0.2137 |
bn2c_branch2b/beta:0 | (64,) | -2.4090 | +1.8192 | +0.6302 |
bn2c_branch2b/moving_mean:0 | (64,) | -1.4737 | +0.0594 | +0.2559 |
bn2c_branch2b/moving_variance:0 | (64,) | +0.2314 | +2.1085 | +0.3072 |
res2c_branch2c/kernel:0 | (1, 1, 64, 256) | -0.2935 | +0.2596 | +0.0434 |
res2c_branch2c/bias:0 | (256,) | -0.0041 | +0.0184 | +0.0029 |
bn2c_branch2c/gamma:0 | (256,) | -0.0217 | +2.3695 | +0.5250 |
bn2c_branch2c/beta:0 | (256,) | -1.6829 | +1.0992 | +0.4280 |
bn2c_branch2c/moving_mean:0 | (256,) | -1.2568 | +0.7135 | +0.2851 |
bn2c_branch2c/moving_variance:0 | (256,) | +0.0010 | +0.5712 | +0.0975 |
res3a_branch2a/kernel:0 | (1, 1, 256, 128) | -0.4997 | +0.6191 | +0.0305 |
res3a_branch2a/bias:0 | (128,) | -0.0025 | +0.0020 | +0.0009 |
bn3a_branch2a/gamma:0 | (128,) | +0.4899 | +1.3306 | +0.1884 |
bn3a_branch2a/beta:0 | (128,) | -1.8391 | +2.5643 | +0.7573 |
bn3a_branch2a/moving_mean:0 | (128,) | -4.0452 | +1.7707 | +0.8690 |
bn3a_branch2a/moving_variance:0 | (128,) | +0.0620 | +7.9964 | +1.2851 |
res3a_branch2b/kernel:0 | (3, 3, 128, 128) | -0.3225 | +0.4509 | +0.0223 |
res3a_branch2b/bias:0 | (128,) | -0.0011 | +0.0019 | +0.0006 |
bn3a_branch2b/gamma:0 | (128,) | +0.4666 | +1.8240 | +0.2182 |
bn3a_branch2b/beta:0 | (128,) | -1.9434 | +1.8963 | +0.7859 |
bn3a_branch2b/moving_mean:0 | (128,) | -5.8993 | +3.3426 | +2.0065 |
bn3a_branch2b/moving_variance:0 | (128,) | +0.0001 | +6.9908 | +0.9404 |
res3a_branch2c/kernel:0 | (1, 1, 128, 512) | -0.4949 | +0.3345 | +0.0283 |
res3a_branch2c/bias:0 | (512,) | -0.0063 | +0.0078 | +0.0013 |
res3a_branch1/kernel:0 | (1, 1, 256, 512) | -0.4556 | +0.6877 | +0.0290 |
res3a_branch1/bias:0 | (512,) | -0.0055 | +0.0039 | +0.0008 |
bn3a_branch2c/gamma:0 | (512,) | -0.0039 | +3.7005 | +0.6168 |
bn3a_branch2c/beta:0 | (512,) | -0.9616 | +1.4438 | +0.3693 |
bn3a_branch2c/moving_mean:0 | (512,) | -1.6188 | +1.3639 | +0.3736 |
bn3a_branch2c/moving_variance:0 | (512,) | +0.0002 | +0.9085 | +0.1065 |
bn3a_branch1/gamma:0 | (512,) | -0.0158 | +2.6945 | +0.4766 |
bn3a_branch1/beta:0 | (512,) | -0.9616 | +1.4437 | +0.3693 |
bn3a_branch1/moving_mean:0 | (512,) | -3.5990 | +2.8529 | +0.7936 |
bn3a_branch1/moving_variance:0 | (512,) | +0.0030 | +6.5634 | +0.6189 |
res3b_branch2a/kernel:0 | (1, 1, 512, 128) | -0.2015 | +0.1914 | +0.0252 |
res3b_branch2a/bias:0 | (128,) | -0.0015 | +0.0020 | +0.0008 |
bn3b_branch2a/gamma:0 | (128,) | +0.5928 | +1.5316 | +0.1783 |
bn3b_branch2a/beta:0 | (128,) | -3.9542 | +0.6799 | +0.6433 |
bn3b_branch2a/moving_mean:0 | (128,) | -2.6765 | +1.1148 | +0.6228 |
bn3b_branch2a/moving_variance:0 | (128,) | +0.2431 | +3.5601 | +0.5766 |
res3b_branch2b/kernel:0 | (3, 3, 128, 128) | -0.2265 | +0.2805 | +0.0240 |
res3b_branch2b/bias:0 | (128,) | -0.0027 | +0.0051 | +0.0013 |
bn3b_branch2b/gamma:0 | (128,) | +0.4900 | +1.4915 | +0.2334 |
bn3b_branch2b/beta:0 | (128,) | -2.4206 | +1.4218 | +0.6774 |
bn3b_branch2b/moving_mean:0 | (128,) | -2.1795 | +1.2802 | +0.4907 |
bn3b_branch2b/moving_variance:0 | (128,) | +0.0892 | +1.4424 | +0.2128 |
res3b_branch2c/kernel:0 | (1, 1, 128, 512) | -0.3113 | +0.4752 | +0.0289 |
res3b_branch2c/bias:0 | (512,) | -0.0061 | +0.0141 | +0.0020 |
bn3b_branch2c/gamma:0 | (512,) | -0.0431 | +2.0087 | +0.4044 |
bn3b_branch2c/beta:0 | (512,) | -1.5772 | +1.1600 | +0.3742 |
bn3b_branch2c/moving_mean:0 | (512,) | -1.0651 | +0.6899 | +0.2256 |
bn3b_branch2c/moving_variance:0 | (512,) | +0.0002 | +0.1858 | +0.0288 |
res3c_branch2a/kernel:0 | (1, 1, 512, 128) | -0.2672 | +0.2625 | +0.0284 |
res3c_branch2a/bias:0 | (128,) | -0.0017 | +0.0035 | +0.0008 |
bn3c_branch2a/gamma:0 | (128,) | +0.5933 | +1.4906 | +0.1891 |
bn3c_branch2a/beta:0 | (128,) | -2.8070 | +0.7289 | +0.5408 |
bn3c_branch2a/moving_mean:0 | (128,) | -3.0308 | +1.6105 | +0.8393 |
bn3c_branch2a/moving_variance:0 | (128,) | +0.2414 | +4.0907 | +0.8334 |
res3c_branch2b/kernel:0 | (3, 3, 128, 128) | -0.2250 | +0.2017 | +0.0233 |
res3c_branch2b/bias:0 | (128,) | -0.0055 | +0.0081 | +0.0022 |
bn3c_branch2b/gamma:0 | (128,) | +0.4480 | +1.5784 | +0.2838 |
bn3c_branch2b/beta:0 | (128,) | -1.4159 | +1.3200 | +0.5555 |
bn3c_branch2b/moving_mean:0 | (128,) | -1.0064 | +0.5542 | +0.2663 |
bn3c_branch2b/moving_variance:0 | (128,) | +0.1152 | +0.8630 | +0.1393 |
res3c_branch2c/kernel:0 | (1, 1, 128, 512) | -0.3069 | +0.3883 | +0.0264 |
res3c_branch2c/bias:0 | (512,) | -0.0075 | +0.0120 | +0.0020 |
bn3c_branch2c/gamma:0 | (512,) | -0.0409 | +1.8960 | +0.3768 |
bn3c_branch2c/beta:0 | (512,) | -1.5428 | +0.8270 | +0.3608 |
bn3c_branch2c/moving_mean:0 | (512,) | -0.8480 | +0.7275 | +0.1809 |
bn3c_branch2c/moving_variance:0 | (512,) | +0.0002 | +0.1614 | +0.0254 |
res3d_branch2a/kernel:0 | (1, 1, 512, 128) | -0.2583 | +0.2893 | +0.0306 |
res3d_branch2a/bias:0 | (128,) | -0.0014 | +0.0018 | +0.0007 |
bn3d_branch2a/gamma:0 | (128,) | +0.6395 | +1.4617 | +0.1923 |
bn3d_branch2a/beta:0 | (128,) | -2.9768 | +0.6138 | +0.6397 |
bn3d_branch2a/moving_mean:0 | (128,) | -3.4373 | +2.0843 | +0.9585 |
bn3d_branch2a/moving_variance:0 | (128,) | +0.0032 | +3.2415 | +0.5276 |
res3d_branch2b/kernel:0 | (3, 3, 128, 128) | -0.1592 | +0.2480 | +0.0237 |
res3d_branch2b/bias:0 | (128,) | -0.0025 | +0.0062 | +0.0017 |
bn3d_branch2b/gamma:0 | (128,) | +0.6485 | +3.2665 | +0.2892 |
bn3d_branch2b/beta:0 | (128,) | -1.6517 | +1.5628 | +0.5605 |
bn3d_branch2b/moving_mean:0 | (128,) | -0.9797 | +0.3526 | +0.2822 |
bn3d_branch2b/moving_variance:0 | (128,) | +0.2176 | +1.4907 | +0.1816 |
res3d_branch2c/kernel:0 | (1, 1, 128, 512) | -0.2404 | +0.3462 | +0.0271 |
res3d_branch2c/bias:0 | (512,) | -0.0042 | +0.0048 | +0.0015 |
bn3d_branch2c/gamma:0 | (512,) | -0.0272 | +1.9218 | +0.5163 |
bn3d_branch2c/beta:0 | (512,) | -1.0555 | +0.9748 | +0.2711 |
bn3d_branch2c/moving_mean:0 | (512,) | -1.1079 | +0.4287 | +0.2314 |
bn3d_branch2c/moving_variance:0 | (512,) | +0.0002 | +0.3496 | +0.0543 |
res4a_branch2a/kernel:0 | (1, 1, 512, 256) | -0.2763 | +0.2629 | +0.0150 |
res4a_branch2a/bias:0 | (256,) | -0.0012 | +0.0013 | +0.0004 |
bn4a_branch2a/gamma:0 | (256,) | +0.4785 | +1.4612 | +0.1493 |
bn4a_branch2a/beta:0 | (256,) | -1.9415 | +1.1200 | +0.3882 |
bn4a_branch2a/moving_mean:0 | (256,) | -3.8936 | +1.1756 | +0.6395 |
bn4a_branch2a/moving_variance:0 | (256,) | +0.0515 | +2.6553 | +0.2808 |
res4a_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1716 | +0.1965 | +0.0106 |
res4a_branch2b/bias:0 | (256,) | -0.0033 | +0.0037 | +0.0007 |
bn4a_branch2b/gamma:0 | (256,) | +0.4768 | +1.5810 | +0.1980 |
bn4a_branch2b/beta:0 | (256,) | -2.5978 | +1.1149 | +0.4805 |
bn4a_branch2b/moving_mean:0 | (256,) | -2.7021 | +2.6603 | +0.5277 |
bn4a_branch2b/moving_variance:0 | (256,) | +0.1003 | +1.1930 | +0.1722 |
res4a_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.2861 | +0.1943 | +0.0141 |
res4a_branch2c/bias:0 | (1024,) | -0.0049 | +0.0115 | +0.0012 |
res4a_branch1/kernel:0 | (1, 1, 512, 1024) | -0.3615 | +0.3428 | +0.0159 |
res4a_branch1/bias:0 | (1024,) | -0.0015 | +0.0015 | +0.0004 |
bn4a_branch2c/gamma:0 | (1024,) | -0.0104 | +2.8173 | +0.4544 |
bn4a_branch2c/beta:0 | (1024,) | -0.5242 | +2.0439 | +0.2862 |
bn4a_branch2c/moving_mean:0 | (1024,) | -0.4020 | +0.2339 | +0.0729 |
bn4a_branch2c/moving_variance:0 | (1024,) | +0.0000 | +0.1119 | +0.0107 |
bn4a_branch1/gamma:0 | (1024,) | +0.1723 | +3.9846 | +0.7125 |
bn4a_branch1/beta:0 | (1024,) | -0.5242 | +2.0441 | +0.2862 |
bn4a_branch1/moving_mean:0 | (1024,) | -4.9091 | +2.9439 | +0.7998 |
bn4a_branch1/moving_variance:0 | (1024,) | +0.0413 | +6.4599 | +0.5613 |
res4b_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1251 | +0.1742 | +0.0082 |
res4b_branch2a/bias:0 | (256,) | -0.0007 | +0.0006 | +0.0002 |
bn4b_branch2a/gamma:0 | (256,) | +0.4161 | +1.5930 | +0.1866 |
bn4b_branch2a/beta:0 | (256,) | -2.2049 | +2.0415 | +0.4853 |
bn4b_branch2a/moving_mean:0 | (256,) | -3.9798 | +2.5647 | +0.9971 |
bn4b_branch2a/moving_variance:0 | (256,) | +0.1146 | +9.1091 | +1.1798 |
res4b_branch2b/kernel:0 | (3, 3, 256, 256) | -0.0968 | +0.1622 | +0.0073 |
res4b_branch2b/bias:0 | (256,) | -0.0022 | +0.0018 | +0.0006 |
bn4b_branch2b/gamma:0 | (256,) | +0.4992 | +1.4646 | +0.1878 |
bn4b_branch2b/beta:0 | (256,) | -1.6821 | +0.5865 | +0.4319 |
bn4b_branch2b/moving_mean:0 | (256,) | -5.5636 | +1.5920 | +0.8391 |
bn4b_branch2b/moving_variance:0 | (256,) | +0.0140 | +1.2341 | +0.1852 |
res4b_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.2025 | +0.2962 | +0.0104 |
res4b_branch2c/bias:0 | (1024,) | -0.0057 | +0.0110 | +0.0017 |
bn4b_branch2c/gamma:0 | (1024,) | -0.0006 | +3.1556 | +0.3625 |
bn4b_branch2c/beta:0 | (1024,) | -1.0586 | +0.9457 | +0.1802 |
bn4b_branch2c/moving_mean:0 | (1024,) | -0.3958 | +0.3607 | +0.0831 |
bn4b_branch2c/moving_variance:0 | (1024,) | +0.0000 | +0.1685 | +0.0150 |
res4c_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1010 | +0.1236 | +0.0083 |
res4c_branch2a/bias:0 | (256,) | -0.0006 | +0.0006 | +0.0002 |
bn4c_branch2a/gamma:0 | (256,) | +0.5716 | +1.7534 | +0.1407 |
bn4c_branch2a/beta:0 | (256,) | -0.9249 | +1.3189 | +0.3732 |
bn4c_branch2a/moving_mean:0 | (256,) | -3.9561 | +1.9207 | +1.0270 |
bn4c_branch2a/moving_variance:0 | (256,) | +0.2736 | +4.1165 | +0.5892 |
res4c_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1008 | +0.1099 | +0.0075 |
res4c_branch2b/bias:0 | (256,) | -0.0015 | +0.0021 | +0.0005 |
bn4c_branch2b/gamma:0 | (256,) | +0.5034 | +1.2173 | +0.1483 |
bn4c_branch2b/beta:0 | (256,) | -1.4417 | +0.5756 | +0.3369 |
bn4c_branch2b/moving_mean:0 | (256,) | -2.9200 | +1.5946 | +0.5483 |
bn4c_branch2b/moving_variance:0 | (256,) | +0.0281 | +1.2198 | +0.1510 |
res4c_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1328 | +0.1666 | +0.0108 |
res4c_branch2c/bias:0 | (1024,) | -0.0057 | +0.0144 | +0.0018 |
bn4c_branch2c/gamma:0 | (1024,) | +0.0043 | +2.2694 | +0.2649 |
bn4c_branch2c/beta:0 | (1024,) | -1.1019 | +0.7349 | +0.1791 |
bn4c_branch2c/moving_mean:0 | (1024,) | -0.3293 | +0.1280 | +0.0515 |
bn4c_branch2c/moving_variance:0 | (1024,) | +0.0001 | +0.0869 | +0.0065 |
res4d_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1169 | +0.1507 | +0.0104 |
res4d_branch2a/bias:0 | (256,) | -0.0010 | +0.0006 | +0.0003 |
bn4d_branch2a/gamma:0 | (256,) | +0.5686 | +1.4401 | +0.1488 |
bn4d_branch2a/beta:0 | (256,) | -1.3452 | +0.4773 | +0.3038 |
bn4d_branch2a/moving_mean:0 | (256,) | -2.9391 | +2.3041 | +0.8307 |
bn4d_branch2a/moving_variance:0 | (256,) | +0.2651 | +4.1963 | +0.5838 |
res4d_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1036 | +0.0993 | +0.0088 |
res4d_branch2b/bias:0 | (256,) | -0.0035 | +0.0054 | +0.0014 |
bn4d_branch2b/gamma:0 | (256,) | +0.4286 | +1.5386 | +0.1594 |
bn4d_branch2b/beta:0 | (256,) | -1.4343 | +0.3851 | +0.2820 |
bn4d_branch2b/moving_mean:0 | (256,) | -1.1160 | +0.5873 | +0.2232 |
bn4d_branch2b/moving_variance:0 | (256,) | +0.0355 | +0.4098 | +0.0649 |
res4d_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.2382 | +0.1296 | +0.0120 |
res4d_branch2c/bias:0 | (1024,) | -0.0070 | +0.0199 | +0.0023 |
bn4d_branch2c/gamma:0 | (1024,) | +0.0461 | +2.8471 | +0.3289 |
bn4d_branch2c/beta:0 | (1024,) | -1.3527 | +0.5924 | +0.2292 |
bn4d_branch2c/moving_mean:0 | (1024,) | -0.2602 | +0.0767 | +0.0430 |
bn4d_branch2c/moving_variance:0 | (1024,) | +0.0013 | +0.0854 | +0.0057 |
res4e_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1474 | +0.1154 | +0.0103 |
res4e_branch2a/bias:0 | (256,) | -0.0006 | +0.0009 | +0.0003 |
bn4e_branch2a/gamma:0 | (256,) | +0.6414 | +1.3680 | +0.1230 |
bn4e_branch2a/beta:0 | (256,) | -1.0867 | +0.3564 | +0.2688 |
bn4e_branch2a/moving_mean:0 | (256,) | -3.8987 | +1.4863 | +1.0202 |
bn4e_branch2a/moving_variance:0 | (256,) | +0.2908 | +4.1538 | +0.5407 |
res4e_branch2b/kernel:0 | (3, 3, 256, 256) | -0.0862 | +0.0939 | +0.0091 |
res4e_branch2b/bias:0 | (256,) | -0.0029 | +0.0051 | +0.0010 |
bn4e_branch2b/gamma:0 | (256,) | +0.5490 | +1.2861 | +0.1311 |
bn4e_branch2b/beta:0 | (256,) | -1.2790 | +0.2216 | +0.2528 |
bn4e_branch2b/moving_mean:0 | (256,) | -1.0716 | +0.6271 | +0.2514 |
bn4e_branch2b/moving_variance:0 | (256,) | +0.0388 | +0.5987 | +0.0750 |
res4e_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.2014 | +0.1777 | +0.0121 |
res4e_branch2c/bias:0 | (1024,) | -0.0053 | +0.0107 | +0.0020 |
bn4e_branch2c/gamma:0 | (1024,) | +0.0251 | +1.7328 | +0.1828 |
bn4e_branch2c/beta:0 | (1024,) | -1.0507 | +0.4076 | +0.1765 |
bn4e_branch2c/moving_mean:0 | (1024,) | -0.2043 | +0.0873 | +0.0359 |
bn4e_branch2c/moving_variance:0 | (1024,) | +0.0007 | +0.0373 | +0.0033 |
res4f_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.0860 | +0.1289 | +0.0106 |
res4f_branch2a/bias:0 | (256,) | -0.0006 | +0.0010 | +0.0003 |
bn4f_branch2a/gamma:0 | (256,) | +0.6962 | +1.3808 | +0.0996 |
bn4f_branch2a/beta:0 | (256,) | -1.1552 | +0.3890 | +0.2462 |
bn4f_branch2a/moving_mean:0 | (256,) | -4.3923 | +2.0997 | +1.0476 |
bn4f_branch2a/moving_variance:0 | (256,) | +0.3235 | +3.9428 | +0.6106 |
res4f_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1023 | +0.1056 | +0.0097 |
res4f_branch2b/bias:0 | (256,) | -0.0038 | +0.0037 | +0.0010 |
bn4f_branch2b/gamma:0 | (256,) | +0.4326 | +1.2403 | +0.1092 |
bn4f_branch2b/beta:0 | (256,) | -1.4296 | +0.5155 | +0.2444 |
bn4f_branch2b/moving_mean:0 | (256,) | -1.6409 | +1.4733 | +0.3083 |
bn4f_branch2b/moving_variance:0 | (256,) | +0.0403 | +0.5428 | +0.0756 |
res4f_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1396 | +0.1322 | +0.0121 |
res4f_branch2c/bias:0 | (1024,) | -0.0073 | +0.0118 | +0.0025 |
bn4f_branch2c/gamma:0 | (1024,) | +0.1776 | +1.6173 | +0.1734 |
bn4f_branch2c/beta:0 | (1024,) | -0.9800 | +0.2974 | +0.1410 |
bn4f_branch2c/moving_mean:0 | (1024,) | -0.1488 | +0.0678 | +0.0309 |
bn4f_branch2c/moving_variance:0 | (1024,) | +0.0008 | +0.0160 | +0.0021 |
res4g_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1154 | +0.2504 | +0.0109 |
res4g_branch2a/bias:0 | (256,) | -0.0008 | +0.0007 | +0.0003 |
bn4g_branch2a/gamma:0 | (256,) | +0.6048 | +1.2029 | +0.1063 |
bn4g_branch2a/beta:0 | (256,) | -1.2676 | +0.2347 | +0.2817 |
bn4g_branch2a/moving_mean:0 | (256,) | -4.1649 | +1.3777 | +0.9897 |
bn4g_branch2a/moving_variance:0 | (256,) | +0.2755 | +3.4157 | +0.6010 |
res4g_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1271 | +0.1231 | +0.0098 |
res4g_branch2b/bias:0 | (256,) | -0.0044 | +0.0032 | +0.0012 |
bn4g_branch2b/gamma:0 | (256,) | +0.4751 | +1.7336 | +0.1333 |
bn4g_branch2b/beta:0 | (256,) | -1.2614 | +0.1900 | +0.2693 |
bn4g_branch2b/moving_mean:0 | (256,) | -0.9138 | +0.8013 | +0.2403 |
bn4g_branch2b/moving_variance:0 | (256,) | +0.0337 | +0.5631 | +0.0711 |
res4g_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1536 | +0.2370 | +0.0120 |
res4g_branch2c/bias:0 | (1024,) | -0.0083 | +0.0083 | +0.0024 |
bn4g_branch2c/gamma:0 | (1024,) | +0.0907 | +1.8097 | +0.1972 |
bn4g_branch2c/beta:0 | (1024,) | -0.9016 | +0.2926 | +0.1411 |
bn4g_branch2c/moving_mean:0 | (1024,) | -0.1636 | +0.0711 | +0.0339 |
bn4g_branch2c/moving_variance:0 | (1024,) | +0.0009 | +0.0321 | +0.0033 |
res4h_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1293 | +0.1603 | +0.0118 |
res4h_branch2a/bias:0 | (256,) | -0.0009 | +0.0007 | +0.0003 |
bn4h_branch2a/gamma:0 | (256,) | +0.6202 | +1.2079 | +0.0980 |
bn4h_branch2a/beta:0 | (256,) | -1.4124 | +0.0712 | +0.2610 |
bn4h_branch2a/moving_mean:0 | (256,) | -3.4425 | +2.5030 | +0.8797 |
bn4h_branch2a/moving_variance:0 | (256,) | +0.3146 | +2.6701 | +0.4704 |
res4h_branch2b/kernel:0 | (3, 3, 256, 256) | -0.0999 | +0.1200 | +0.0103 |
res4h_branch2b/bias:0 | (256,) | -0.0056 | +0.0069 | +0.0016 |
bn4h_branch2b/gamma:0 | (256,) | +0.4724 | +1.4768 | +0.1470 |
bn4h_branch2b/beta:0 | (256,) | -1.5817 | +0.4324 | +0.2521 |
bn4h_branch2b/moving_mean:0 | (256,) | -0.6978 | +0.9014 | +0.1644 |
bn4h_branch2b/moving_variance:0 | (256,) | +0.0297 | +0.4195 | +0.0563 |
res4h_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1493 | +0.2391 | +0.0123 |
res4h_branch2c/bias:0 | (1024,) | -0.0071 | +0.0081 | +0.0024 |
bn4h_branch2c/gamma:0 | (1024,) | +0.0532 | +2.4672 | +0.2958 |
bn4h_branch2c/beta:0 | (1024,) | -0.7953 | +0.3313 | +0.1519 |
bn4h_branch2c/moving_mean:0 | (1024,) | -0.1890 | +0.1145 | +0.0347 |
bn4h_branch2c/moving_variance:0 | (1024,) | +0.0006 | +0.0432 | +0.0041 |
res4i_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1308 | +0.2974 | +0.0132 |
res4i_branch2a/bias:0 | (256,) | -0.0008 | +0.0009 | +0.0003 |
bn4i_branch2a/gamma:0 | (256,) | +0.3549 | +1.0498 | +0.1261 |
bn4i_branch2a/beta:0 | (256,) | -1.4706 | +0.4588 | +0.2771 |
bn4i_branch2a/moving_mean:0 | (256,) | -3.8994 | +2.5413 | +1.0756 |
bn4i_branch2a/moving_variance:0 | (256,) | +0.6003 | +6.0203 | +0.5194 |
res4i_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1302 | +0.1541 | +0.0093 |
res4i_branch2b/bias:0 | (256,) | -0.0063 | +0.0112 | +0.0028 |
bn4i_branch2b/gamma:0 | (256,) | +0.5664 | +1.6229 | +0.1404 |
bn4i_branch2b/beta:0 | (256,) | -1.4855 | +0.4776 | +0.2638 |
bn4i_branch2b/moving_mean:0 | (256,) | -0.3930 | +0.1225 | +0.0744 |
bn4i_branch2b/moving_variance:0 | (256,) | +0.0113 | +0.0744 | +0.0117 |
res4i_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1554 | +0.1568 | +0.0120 |
res4i_branch2c/bias:0 | (1024,) | -0.0069 | +0.0131 | +0.0017 |
bn4i_branch2c/gamma:0 | (1024,) | +0.0400 | +2.1618 | +0.1852 |
bn4i_branch2c/beta:0 | (1024,) | -0.5914 | +0.7033 | +0.1253 |
bn4i_branch2c/moving_mean:0 | (1024,) | -0.3092 | +0.1096 | +0.0572 |
bn4i_branch2c/moving_variance:0 | (1024,) | +0.0007 | +0.0708 | +0.0049 |
res4j_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1264 | +0.1794 | +0.0126 |
res4j_branch2a/bias:0 | (256,) | -0.0010 | +0.0006 | +0.0003 |
bn4j_branch2a/gamma:0 | (256,) | +0.5069 | +1.2823 | +0.1149 |
bn4j_branch2a/beta:0 | (256,) | -1.9055 | +0.2154 | +0.2870 |
bn4j_branch2a/moving_mean:0 | (256,) | -4.5494 | +1.8513 | +0.9530 |
bn4j_branch2a/moving_variance:0 | (256,) | +0.5073 | +5.0598 | +0.4980 |
res4j_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1036 | +0.2351 | +0.0103 |
res4j_branch2b/bias:0 | (256,) | -0.0067 | +0.0059 | +0.0020 |
bn4j_branch2b/gamma:0 | (256,) | +0.4131 | +1.4478 | +0.1323 |
bn4j_branch2b/beta:0 | (256,) | -1.9463 | +0.4984 | +0.2980 |
bn4j_branch2b/moving_mean:0 | (256,) | -0.7600 | +0.4444 | +0.1658 |
bn4j_branch2b/moving_variance:0 | (256,) | +0.0380 | +0.3044 | +0.0348 |
res4j_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1406 | +0.1723 | +0.0121 |
res4j_branch2c/bias:0 | (1024,) | -0.0062 | +0.0091 | +0.0025 |
bn4j_branch2c/gamma:0 | (1024,) | +0.0288 | +2.0876 | +0.1976 |
bn4j_branch2c/beta:0 | (1024,) | -0.8279 | +0.1712 | +0.1242 |
bn4j_branch2c/moving_mean:0 | (1024,) | -0.1671 | +0.0815 | +0.0318 |
bn4j_branch2c/moving_variance:0 | (1024,) | +0.0002 | +0.0240 | +0.0027 |
res4k_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1369 | +0.1895 | +0.0115 |
res4k_branch2a/bias:0 | (256,) | -0.0009 | +0.0007 | +0.0003 |
bn4k_branch2a/gamma:0 | (256,) | +0.5384 | +1.1987 | +0.1222 |
bn4k_branch2a/beta:0 | (256,) | -1.7274 | +0.3939 | +0.3094 |
bn4k_branch2a/moving_mean:0 | (256,) | -5.5573 | +2.2690 | +1.0985 |
bn4k_branch2a/moving_variance:0 | (256,) | +0.3051 | +3.5934 | +0.4903 |
res4k_branch2b/kernel:0 | (3, 3, 256, 256) | -0.0799 | +0.1296 | +0.0095 |
res4k_branch2b/bias:0 | (256,) | -0.0060 | +0.0041 | +0.0014 |
bn4k_branch2b/gamma:0 | (256,) | +0.4960 | +1.2266 | +0.1243 |
bn4k_branch2b/beta:0 | (256,) | -1.2792 | +0.2008 | +0.2593 |
bn4k_branch2b/moving_mean:0 | (256,) | -0.8575 | +1.2273 | +0.2591 |
bn4k_branch2b/moving_variance:0 | (256,) | +0.0168 | +0.3693 | +0.0569 |
res4k_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1247 | +0.2241 | +0.0114 |
res4k_branch2c/bias:0 | (1024,) | -0.0067 | +0.0081 | +0.0024 |
bn4k_branch2c/gamma:0 | (1024,) | +0.1148 | +1.9941 | +0.2081 |
bn4k_branch2c/beta:0 | (1024,) | -1.6103 | +0.1858 | +0.1599 |
bn4k_branch2c/moving_mean:0 | (1024,) | -0.1446 | +0.0679 | +0.0319 |
bn4k_branch2c/moving_variance:0 | (1024,) | +0.0010 | +0.0395 | +0.0042 |
res4l_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.2041 | +0.1931 | +0.0135 |
res4l_branch2a/bias:0 | (256,) | -0.0011 | +0.0008 | +0.0003 |
bn4l_branch2a/gamma:0 | (256,) | +0.4169 | +1.5267 | +0.1211 |
bn4l_branch2a/beta:0 | (256,) | -1.8435 | +0.3071 | +0.2976 |
bn4l_branch2a/moving_mean:0 | (256,) | -4.0608 | +2.0131 | +1.0383 |
bn4l_branch2a/moving_variance:0 | (256,) | +0.5797 | +5.8934 | +0.5997 |
res4l_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1075 | +0.1778 | +0.0101 |
res4l_branch2b/bias:0 | (256,) | -0.0089 | +0.0092 | +0.0025 |
bn4l_branch2b/gamma:0 | (256,) | +0.4045 | +1.3411 | +0.1290 |
bn4l_branch2b/beta:0 | (256,) | -1.4767 | +0.4854 | +0.2710 |
bn4l_branch2b/moving_mean:0 | (256,) | -0.3820 | +0.2590 | +0.0978 |
bn4l_branch2b/moving_variance:0 | (256,) | +0.0194 | +0.1928 | +0.0195 |
res4l_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1198 | +0.1714 | +0.0123 |
res4l_branch2c/bias:0 | (1024,) | -0.0098 | +0.0088 | +0.0023 |
bn4l_branch2c/gamma:0 | (1024,) | +0.0792 | +1.6725 | +0.1760 |
bn4l_branch2c/beta:0 | (1024,) | -1.0323 | +0.7015 | +0.1526 |
bn4l_branch2c/moving_mean:0 | (1024,) | -0.1346 | +0.0818 | +0.0317 |
bn4l_branch2c/moving_variance:0 | (1024,) | +0.0007 | +0.0367 | +0.0027 |
res4m_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.0814 | +0.1539 | +0.0118 |
res4m_branch2a/bias:0 | (256,) | -0.0008 | +0.0006 | +0.0003 |
bn4m_branch2a/gamma:0 | (256,) | +0.5300 | +1.1776 | +0.0988 |
bn4m_branch2a/beta:0 | (256,) | -1.2997 | +0.3071 | +0.2241 |
bn4m_branch2a/moving_mean:0 | (256,) | -6.0726 | +1.2736 | +1.0509 |
bn4m_branch2a/moving_variance:0 | (256,) | +0.4134 | +5.9330 | +0.5904 |
res4m_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1037 | +0.1375 | +0.0091 |
res4m_branch2b/bias:0 | (256,) | -0.0072 | +0.0071 | +0.0020 |
bn4m_branch2b/gamma:0 | (256,) | +0.5814 | +1.1881 | +0.1004 |
bn4m_branch2b/beta:0 | (256,) | -1.3691 | +0.1966 | +0.2357 |
bn4m_branch2b/moving_mean:0 | (256,) | -0.6768 | +0.5120 | +0.1552 |
bn4m_branch2b/moving_variance:0 | (256,) | +0.0234 | +0.3098 | +0.0412 |
res4m_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1469 | +0.1552 | +0.0113 |
res4m_branch2c/bias:0 | (1024,) | -0.0083 | +0.0104 | +0.0025 |
bn4m_branch2c/gamma:0 | (1024,) | +0.1858 | +1.7955 | +0.1699 |
bn4m_branch2c/beta:0 | (1024,) | -0.7632 | +0.5809 | +0.1474 |
bn4m_branch2c/moving_mean:0 | (1024,) | -0.1689 | +0.0692 | +0.0349 |
bn4m_branch2c/moving_variance:0 | (1024,) | +0.0007 | +0.0343 | +0.0037 |
res4n_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1217 | +0.1600 | +0.0128 |
res4n_branch2a/bias:0 | (256,) | -0.0012 | +0.0006 | +0.0003 |
bn4n_branch2a/gamma:0 | (256,) | +0.4676 | +1.0852 | +0.1114 |
bn4n_branch2a/beta:0 | (256,) | -1.2727 | +0.0440 | +0.2325 |
bn4n_branch2a/moving_mean:0 | (256,) | -3.7141 | +1.9744 | +0.9055 |
bn4n_branch2a/moving_variance:0 | (256,) | +0.4856 | +2.9867 | +0.3987 |
res4n_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1221 | +0.1551 | +0.0089 |
res4n_branch2b/bias:0 | (256,) | -0.0099 | +0.0089 | +0.0028 |
bn4n_branch2b/gamma:0 | (256,) | +0.4873 | +1.1907 | +0.1089 |
bn4n_branch2b/beta:0 | (256,) | -1.0325 | +0.5976 | +0.2187 |
bn4n_branch2b/moving_mean:0 | (256,) | -0.3456 | +0.0662 | +0.0728 |
bn4n_branch2b/moving_variance:0 | (256,) | +0.0111 | +0.2385 | +0.0187 |
res4n_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.0992 | +0.1609 | +0.0109 |
res4n_branch2c/bias:0 | (1024,) | -0.0078 | +0.0087 | +0.0021 |
bn4n_branch2c/gamma:0 | (1024,) | +0.1917 | +1.6763 | +0.1232 |
bn4n_branch2c/beta:0 | (1024,) | -0.7562 | +0.6426 | +0.1316 |
bn4n_branch2c/moving_mean:0 | (1024,) | -0.2036 | +0.1072 | +0.0411 |
bn4n_branch2c/moving_variance:0 | (1024,) | +0.0013 | +0.0381 | +0.0031 |
res4o_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.0879 | +0.1375 | +0.0125 |
res4o_branch2a/bias:0 | (256,) | -0.0009 | +0.0009 | +0.0003 |
bn4o_branch2a/gamma:0 | (256,) | +0.4154 | +1.0786 | +0.1032 |
bn4o_branch2a/beta:0 | (256,) | -1.5070 | +0.1578 | +0.2357 |
bn4o_branch2a/moving_mean:0 | (256,) | -6.4399 | +2.0957 | +1.2424 |
bn4o_branch2a/moving_variance:0 | (256,) | +0.5261 | +5.2096 | +0.5908 |
res4o_branch2b/kernel:0 | (3, 3, 256, 256) | -0.0939 | +0.1264 | +0.0090 |
res4o_branch2b/bias:0 | (256,) | -0.0093 | +0.0069 | +0.0027 |
bn4o_branch2b/gamma:0 | (256,) | +0.5379 | +1.2134 | +0.1078 |
bn4o_branch2b/beta:0 | (256,) | -1.2517 | +0.4139 | +0.2248 |
bn4o_branch2b/moving_mean:0 | (256,) | -0.3864 | +0.3207 | +0.0934 |
bn4o_branch2b/moving_variance:0 | (256,) | +0.0164 | +0.1712 | +0.0202 |
res4o_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1602 | +0.1698 | +0.0111 |
res4o_branch2c/bias:0 | (1024,) | -0.0073 | +0.0096 | +0.0021 |
bn4o_branch2c/gamma:0 | (1024,) | +0.2356 | +1.9104 | +0.1531 |
bn4o_branch2c/beta:0 | (1024,) | -0.8014 | +0.5613 | +0.1387 |
bn4o_branch2c/moving_mean:0 | (1024,) | -0.2040 | +0.0855 | +0.0429 |
bn4o_branch2c/moving_variance:0 | (1024,) | +0.0009 | +0.0544 | +0.0049 |
res4p_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1453 | +0.2050 | +0.0138 |
res4p_branch2a/bias:0 | (256,) | -0.0008 | +0.0009 | +0.0003 |
bn4p_branch2a/gamma:0 | (256,) | +0.5041 | +1.0460 | +0.0900 |
bn4p_branch2a/beta:0 | (256,) | -1.4744 | +0.0466 | +0.2374 |
bn4p_branch2a/moving_mean:0 | (256,) | -3.5993 | +2.5332 | +1.0418 |
bn4p_branch2a/moving_variance:0 | (256,) | +0.6268 | +3.2764 | +0.5098 |
res4p_branch2b/kernel:0 | (3, 3, 256, 256) | -0.0963 | +0.1146 | +0.0102 |
res4p_branch2b/bias:0 | (256,) | -0.0117 | +0.0087 | +0.0026 |
bn4p_branch2b/gamma:0 | (256,) | +0.4508 | +1.3897 | +0.1299 |
bn4p_branch2b/beta:0 | (256,) | -1.4155 | +0.4056 | +0.2478 |
bn4p_branch2b/moving_mean:0 | (256,) | -0.2807 | +0.1532 | +0.0755 |
bn4p_branch2b/moving_variance:0 | (256,) | +0.0209 | +0.1309 | +0.0195 |
res4p_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1161 | +0.1738 | +0.0122 |
res4p_branch2c/bias:0 | (1024,) | -0.0087 | +0.0081 | +0.0020 |
bn4p_branch2c/gamma:0 | (1024,) | +0.1803 | +1.7117 | +0.1949 |
bn4p_branch2c/beta:0 | (1024,) | -1.0347 | +0.3854 | +0.1619 |
bn4p_branch2c/moving_mean:0 | (1024,) | -0.1642 | +0.0812 | +0.0336 |
bn4p_branch2c/moving_variance:0 | (1024,) | +0.0010 | +0.0413 | +0.0038 |
res4q_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1236 | +0.2559 | +0.0137 |
res4q_branch2a/bias:0 | (256,) | -0.0008 | +0.0012 | +0.0003 |
bn4q_branch2a/gamma:0 | (256,) | +0.3504 | +1.0037 | +0.1050 |
bn4q_branch2a/beta:0 | (256,) | -1.5841 | +0.3542 | +0.2878 |
bn4q_branch2a/moving_mean:0 | (256,) | -5.4757 | +2.7636 | +1.1594 |
bn4q_branch2a/moving_variance:0 | (256,) | +0.4812 | +10.5219 | +0.8778 |
res4q_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1804 | +0.2048 | +0.0089 |
res4q_branch2b/bias:0 | (256,) | -0.0106 | +0.0080 | +0.0028 |
bn4q_branch2b/gamma:0 | (256,) | +0.6510 | +1.4631 | +0.1185 |
bn4q_branch2b/beta:0 | (256,) | -1.1869 | +0.3730 | +0.2479 |
bn4q_branch2b/moving_mean:0 | (256,) | -0.2944 | +0.1277 | +0.0664 |
bn4q_branch2b/moving_variance:0 | (256,) | +0.0110 | +0.0925 | +0.0102 |
res4q_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1754 | +0.1839 | +0.0118 |
res4q_branch2c/bias:0 | (1024,) | -0.0073 | +0.0053 | +0.0015 |
bn4q_branch2c/gamma:0 | (1024,) | +0.0368 | +2.1137 | +0.2127 |
bn4q_branch2c/beta:0 | (1024,) | -0.7801 | +0.3531 | +0.1493 |
bn4q_branch2c/moving_mean:0 | (1024,) | -0.3336 | +0.1393 | +0.0609 |
bn4q_branch2c/moving_variance:0 | (1024,) | +0.0006 | +0.0498 | +0.0058 |
res4r_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1730 | +0.2589 | +0.0137 |
res4r_branch2a/bias:0 | (256,) | -0.0009 | +0.0010 | +0.0003 |
bn4r_branch2a/gamma:0 | (256,) | +0.2862 | +0.9191 | +0.1058 |
bn4r_branch2a/beta:0 | (256,) | -1.3459 | +0.2720 | +0.2652 |
bn4r_branch2a/moving_mean:0 | (256,) | -2.5019 | +3.6722 | +1.0108 |
bn4r_branch2a/moving_variance:0 | (256,) | +0.6803 | +5.8562 | +0.5767 |
res4r_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1322 | +0.1870 | +0.0086 |
res4r_branch2b/bias:0 | (256,) | -0.0092 | +0.0095 | +0.0031 |
bn4r_branch2b/gamma:0 | (256,) | +0.5149 | +1.4533 | +0.1367 |
bn4r_branch2b/beta:0 | (256,) | -0.9097 | +0.6818 | +0.2112 |
bn4r_branch2b/moving_mean:0 | (256,) | -0.2223 | +0.1066 | +0.0609 |
bn4r_branch2b/moving_variance:0 | (256,) | +0.0062 | +0.0611 | +0.0083 |
res4r_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.0890 | +0.1848 | +0.0112 |
res4r_branch2c/bias:0 | (1024,) | -0.0115 | +0.0075 | +0.0017 |
bn4r_branch2c/gamma:0 | (1024,) | +0.1361 | +1.7486 | +0.1512 |
bn4r_branch2c/beta:0 | (1024,) | -0.7476 | +0.3266 | +0.1342 |
bn4r_branch2c/moving_mean:0 | (1024,) | -0.2405 | +0.1581 | +0.0556 |
bn4r_branch2c/moving_variance:0 | (1024,) | +0.0019 | +0.0666 | +0.0043 |
res4s_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1326 | +0.1900 | +0.0130 |
res4s_branch2a/bias:0 | (256,) | -0.0007 | +0.0008 | +0.0003 |
bn4s_branch2a/gamma:0 | (256,) | +0.3468 | +0.9720 | +0.1053 |
bn4s_branch2a/beta:0 | (256,) | -1.2850 | +0.5307 | +0.2683 |
bn4s_branch2a/moving_mean:0 | (256,) | -9.8440 | +2.2720 | +1.2494 |
bn4s_branch2a/moving_variance:0 | (256,) | +0.5810 | +17.3876 | +1.1579 |
res4s_branch2b/kernel:0 | (3, 3, 256, 256) | -0.2008 | +0.1875 | +0.0085 |
res4s_branch2b/bias:0 | (256,) | -0.0109 | +0.0088 | +0.0028 |
bn4s_branch2b/gamma:0 | (256,) | +0.5201 | +1.1966 | +0.1040 |
bn4s_branch2b/beta:0 | (256,) | -1.1253 | +0.3305 | +0.2241 |
bn4s_branch2b/moving_mean:0 | (256,) | -0.3961 | +0.1227 | +0.0815 |
bn4s_branch2b/moving_variance:0 | (256,) | +0.0097 | +0.0778 | +0.0102 |
res4s_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1133 | +0.1737 | +0.0112 |
res4s_branch2c/bias:0 | (1024,) | -0.0104 | +0.0076 | +0.0019 |
bn4s_branch2c/gamma:0 | (1024,) | +0.1482 | +1.7350 | +0.1424 |
bn4s_branch2c/beta:0 | (1024,) | -0.7700 | +0.4369 | +0.1229 |
bn4s_branch2c/moving_mean:0 | (1024,) | -0.1928 | +0.1375 | +0.0475 |
bn4s_branch2c/moving_variance:0 | (1024,) | +0.0015 | +0.0487 | +0.0035 |
res4t_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1664 | +0.1841 | +0.0131 |
res4t_branch2a/bias:0 | (256,) | -0.0013 | +0.0016 | +0.0004 |
bn4t_branch2a/gamma:0 | (256,) | +0.4422 | +1.2399 | +0.1050 |
bn4t_branch2a/beta:0 | (256,) | -1.1249 | +0.2680 | +0.2436 |
bn4t_branch2a/moving_mean:0 | (256,) | -6.2349 | +2.5628 | +1.3091 |
bn4t_branch2a/moving_variance:0 | (256,) | +0.4907 | +5.1029 | +0.5986 |
res4t_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1324 | +0.1087 | +0.0094 |
res4t_branch2b/bias:0 | (256,) | -0.0096 | +0.0074 | +0.0024 |
bn4t_branch2b/gamma:0 | (256,) | +0.4694 | +1.2439 | +0.1069 |
bn4t_branch2b/beta:0 | (256,) | -1.1027 | +0.6878 | +0.2195 |
bn4t_branch2b/moving_mean:0 | (256,) | -0.6517 | +0.2553 | +0.1409 |
bn4t_branch2b/moving_variance:0 | (256,) | +0.0288 | +0.2597 | +0.0352 |
res4t_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1675 | +0.1682 | +0.0118 |
res4t_branch2c/bias:0 | (1024,) | -0.0144 | +0.0073 | +0.0019 |
bn4t_branch2c/gamma:0 | (1024,) | +0.2195 | +1.9902 | +0.1718 |
bn4t_branch2c/beta:0 | (1024,) | -0.7889 | +0.3014 | +0.1334 |
bn4t_branch2c/moving_mean:0 | (1024,) | -0.1941 | +0.1983 | +0.0427 |
bn4t_branch2c/moving_variance:0 | (1024,) | +0.0010 | +0.0383 | +0.0034 |
res4u_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1064 | +0.1593 | +0.0122 |
res4u_branch2a/bias:0 | (256,) | -0.0010 | +0.0009 | +0.0003 |
bn4u_branch2a/gamma:0 | (256,) | +0.2925 | +1.1103 | +0.1152 |
bn4u_branch2a/beta:0 | (256,) | -1.2993 | +0.5482 | +0.2517 |
bn4u_branch2a/moving_mean:0 | (256,) | -8.9484 | +3.0622 | +1.2511 |
bn4u_branch2a/moving_variance:0 | (256,) | +0.3403 | +10.5238 | +1.0425 |
res4u_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1250 | +0.1247 | +0.0080 |
res4u_branch2b/bias:0 | (256,) | -0.0125 | +0.0058 | +0.0024 |
bn4u_branch2b/gamma:0 | (256,) | +0.6188 | +1.3505 | +0.1055 |
bn4u_branch2b/beta:0 | (256,) | -0.9702 | +0.4346 | +0.2031 |
bn4u_branch2b/moving_mean:0 | (256,) | -0.4856 | +0.3687 | +0.1192 |
bn4u_branch2b/moving_variance:0 | (256,) | +0.0157 | +0.1070 | +0.0150 |
res4u_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1426 | +0.2122 | +0.0106 |
res4u_branch2c/bias:0 | (1024,) | -0.0084 | +0.0064 | +0.0016 |
bn4u_branch2c/gamma:0 | (1024,) | +0.0975 | +1.7799 | +0.1396 |
bn4u_branch2c/beta:0 | (1024,) | -0.7753 | +0.6956 | +0.1624 |
bn4u_branch2c/moving_mean:0 | (1024,) | -0.2670 | +0.1501 | +0.0620 |
bn4u_branch2c/moving_variance:0 | (1024,) | +0.0017 | +0.0830 | +0.0062 |
res4v_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1540 | +0.2322 | +0.0125 |
res4v_branch2a/bias:0 | (256,) | -0.0011 | +0.0015 | +0.0003 |
bn4v_branch2a/gamma:0 | (256,) | +0.3923 | +1.0099 | +0.0927 |
bn4v_branch2a/beta:0 | (256,) | -1.2259 | +0.4520 | +0.2694 |
bn4v_branch2a/moving_mean:0 | (256,) | -5.8968 | +2.1904 | +1.1998 |
bn4v_branch2a/moving_variance:0 | (256,) | +0.5714 | +5.2581 | +0.6591 |
res4v_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1413 | +0.1647 | +0.0085 |
res4v_branch2b/bias:0 | (256,) | -0.0130 | +0.0072 | +0.0022 |
bn4v_branch2b/gamma:0 | (256,) | +0.6167 | +1.1526 | +0.0908 |
bn4v_branch2b/beta:0 | (256,) | -1.0165 | +0.8713 | +0.1890 |
bn4v_branch2b/moving_mean:0 | (256,) | -0.4301 | +0.1983 | +0.1069 |
bn4v_branch2b/moving_variance:0 | (256,) | +0.0213 | +0.2294 | +0.0224 |
res4v_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1078 | +0.1724 | +0.0112 |
res4v_branch2c/bias:0 | (1024,) | -0.0045 | +0.0057 | +0.0016 |
bn4v_branch2c/gamma:0 | (1024,) | +0.2332 | +1.6640 | +0.1350 |
bn4v_branch2c/beta:0 | (1024,) | -0.9275 | +0.5133 | +0.1990 |
bn4v_branch2c/moving_mean:0 | (1024,) | -0.2657 | +0.2240 | +0.0562 |
bn4v_branch2c/moving_variance:0 | (1024,) | +0.0014 | +0.0536 | +0.0044 |
res4w_branch2a/kernel:0 | (1, 1, 1024, 256) | -0.1421 | +0.2230 | +0.0128 |
res4w_branch2a/bias:0 | (256,) | -0.0011 | +0.0017 | +0.0003 |
bn4w_branch2a/gamma:0 | (256,) | +0.2562 | +1.0847 | +0.1115 |
bn4w_branch2a/beta:0 | (256,) | -1.4639 | +0.3603 | +0.2947 |
bn4w_branch2a/moving_mean:0 | (256,) | -13.4450 | +3.0168 | +1.9482 |
bn4w_branch2a/moving_variance:0 | (256,) | +0.5124 | +13.2866 | +1.0325 |
res4w_branch2b/kernel:0 | (3, 3, 256, 256) | -0.1053 | +0.1691 | +0.0084 |
res4w_branch2b/bias:0 | (256,) | -0.0078 | +0.0080 | +0.0024 |
bn4w_branch2b/gamma:0 | (256,) | +0.7056 | +1.4043 | +0.0986 |
bn4w_branch2b/beta:0 | (256,) | -0.9674 | +0.3868 | +0.2011 |
bn4w_branch2b/moving_mean:0 | (256,) | -0.2898 | +0.2128 | +0.0745 |
bn4w_branch2b/moving_variance:0 | (256,) | +0.0105 | +0.1042 | +0.0124 |
res4w_branch2c/kernel:0 | (1, 1, 256, 1024) | -0.1479 | +0.1984 | +0.0111 |
res4w_branch2c/bias:0 | (1024,) | -0.0042 | +0.0050 | +0.0014 |
bn4w_branch2c/gamma:0 | (1024,) | +0.0221 | +1.5448 | +0.1517 |
bn4w_branch2c/beta:0 | (1024,) | -0.8512 | +0.5036 | +0.1815 |
bn4w_branch2c/moving_mean:0 | (1024,) | -0.3939 | +0.1908 | +0.1037 |
bn4w_branch2c/moving_variance:0 | (1024,) | +0.0007 | +0.1042 | +0.0077 |
res5a_branch2a/kernel:0 | (1, 1, 1024, 512) | -0.1763 | +0.2315 | +0.0143 |
res5a_branch2a/bias:0 | (512,) | -0.0014 | +0.0011 | +0.0003 |
bn5a_branch2a/gamma:0 | (512,) | +0.5024 | +1.2280 | +0.1224 |
bn5a_branch2a/beta:0 | (512,) | -1.4505 | +0.4926 | +0.3033 |
bn5a_branch2a/moving_mean:0 | (512,) | -13.2482 | +6.5317 | +1.7947 |
bn5a_branch2a/moving_variance:0 | (512,) | +0.8438 | +12.3391 | +1.0853 |
res5a_branch2b/kernel:0 | (3, 3, 512, 512) | -0.2433 | +0.3231 | +0.0091 |
res5a_branch2b/bias:0 | (512,) | -0.0018 | +0.0043 | +0.0008 |
bn5a_branch2b/gamma:0 | (512,) | +0.3045 | +1.4122 | +0.1359 |
bn5a_branch2b/beta:0 | (512,) | -1.6459 | +0.7115 | +0.3145 |
bn5a_branch2b/moving_mean:0 | (512,) | -1.7575 | +1.3904 | +0.2868 |
bn5a_branch2b/moving_variance:0 | (512,) | +0.0838 | +0.8812 | +0.0821 |
res5a_branch2c/kernel:0 | (1, 1, 512, 2048) | -0.2880 | +0.3244 | +0.0122 |
res5a_branch2c/bias:0 | (2048,) | -0.0106 | +0.0227 | +0.0013 |
res5a_branch1/kernel:0 | (1, 1, 1024, 2048) | -0.3702 | +0.4639 | +0.0105 |
res5a_branch1/bias:0 | (2048,) | -0.0008 | +0.0019 | +0.0002 |
bn5a_branch2c/gamma:0 | (2048,) | +0.6632 | +2.6859 | +0.2373 |
bn5a_branch2c/beta:0 | (2048,) | -1.8489 | +1.4639 | +0.2354 |
bn5a_branch2c/moving_mean:0 | (2048,) | -0.4234 | +0.5529 | +0.0570 |
bn5a_branch2c/moving_variance:0 | (2048,) | +0.0023 | +0.1430 | +0.0061 |
bn5a_branch1/gamma:0 | (2048,) | +0.8580 | +4.8497 | +0.5097 |
bn5a_branch1/beta:0 | (2048,) | -1.8488 | +1.4641 | +0.2354 |
bn5a_branch1/moving_mean:0 | (2048,) | -8.1453 | +5.4726 | +1.0711 |
bn5a_branch1/moving_variance:0 | (2048,) | +0.2737 | +5.2058 | +0.3888 |
res5b_branch2a/kernel:0 | (1, 1, 2048, 512) | -0.1497 | +0.2567 | +0.0107 |
res5b_branch2a/bias:0 | (512,) | -0.0015 | +0.0040 | +0.0004 |
bn5b_branch2a/gamma:0 | (512,) | +0.3823 | +1.1338 | +0.0958 |
bn5b_branch2a/beta:0 | (512,) | -1.1818 | +0.6068 | +0.1976 |
bn5b_branch2a/moving_mean:0 | (512,) | -3.0110 | +4.9119 | +0.5962 |
bn5b_branch2a/moving_variance:0 | (512,) | +0.5778 | +4.9932 | +0.4898 |
res5b_branch2b/kernel:0 | (3, 3, 512, 512) | -0.1338 | +0.2394 | +0.0079 |
res5b_branch2b/bias:0 | (512,) | -0.0105 | +0.0073 | +0.0013 |
bn5b_branch2b/gamma:0 | (512,) | +0.5321 | +1.1694 | +0.1049 |
bn5b_branch2b/beta:0 | (512,) | -1.8375 | +0.5450 | +0.2784 |
bn5b_branch2b/moving_mean:0 | (512,) | -1.2009 | +1.5489 | +0.2116 |
bn5b_branch2b/moving_variance:0 | (512,) | +0.0530 | +0.6918 | +0.0500 |
res5b_branch2c/kernel:0 | (1, 1, 512, 2048) | -0.1345 | +0.1962 | +0.0106 |
res5b_branch2c/bias:0 | (2048,) | -0.0181 | +0.0196 | +0.0018 |
bn5b_branch2c/gamma:0 | (2048,) | +0.5622 | +2.4065 | +0.2234 |
bn5b_branch2c/beta:0 | (2048,) | -2.3543 | +0.1656 | +0.2122 |
bn5b_branch2c/moving_mean:0 | (2048,) | -0.2994 | +0.9054 | +0.0436 |
bn5b_branch2c/moving_variance:0 | (2048,) | +0.0017 | +0.1444 | +0.0042 |
res5c_branch2a/kernel:0 | (1, 1, 2048, 512) | -0.1803 | +0.3580 | +0.0115 |
res5c_branch2a/bias:0 | (512,) | -0.0037 | +0.0055 | +0.0005 |
bn5c_branch2a/gamma:0 | (512,) | +0.1743 | +1.1331 | +0.0973 |
bn5c_branch2a/beta:0 | (512,) | -1.3940 | +0.9286 | +0.2532 |
bn5c_branch2a/moving_mean:0 | (512,) | -1.6788 | +4.0125 | +0.4136 |
bn5c_branch2a/moving_variance:0 | (512,) | +0.4097 | +6.2837 | +0.4842 |
res5c_branch2b/kernel:0 | (3, 3, 512, 512) | -0.0940 | +0.0945 | +0.0071 |
res5c_branch2b/bias:0 | (512,) | -0.0092 | +0.0111 | +0.0019 |
bn5c_branch2b/gamma:0 | (512,) | +0.4880 | +1.1432 | +0.0915 |
bn5c_branch2b/beta:0 | (512,) | -1.4251 | +0.3417 | +0.2747 |
bn5c_branch2b/moving_mean:0 | (512,) | -0.6788 | +0.0926 | +0.1057 |
bn5c_branch2b/moving_variance:0 | (512,) | +0.0341 | +0.2969 | +0.0293 |
res5c_branch2c/kernel:0 | (1, 1, 512, 2048) | -0.1305 | +0.1288 | +0.0103 |
res5c_branch2c/bias:0 | (2048,) | -0.0031 | +0.0070 | +0.0011 |
bn5c_branch2c/gamma:0 | (2048,) | +0.5999 | +2.5360 | +0.2191 |
bn5c_branch2c/beta:0 | (2048,) | -4.0101 | -0.6658 | +0.2211 |
bn5c_branch2c/moving_mean:0 | (2048,) | -0.2560 | +0.1848 | +0.0335 |
bn5c_branch2c/moving_variance:0 | (2048,) | +0.0021 | +0.0345 | +0.0027 |
fpn_c5p5/kernel:0 | (1, 1, 2048, 256) | -0.0663 | +0.0641 | +0.0080 |
fpn_c5p5/bias:0 | (256,) | -0.0149 | +0.0132 | +0.0053 |
fpn_c4p4/kernel:0 | (1, 1, 1024, 256) | -0.1159 | +0.0827 | +0.0101 |
fpn_c4p4/bias:0 | (256,) | -0.0049 | +0.0042 | +0.0015 |
fpn_c3p3/kernel:0 | (1, 1, 512, 256) | -0.0507 | +0.0581 | +0.0076 |
fpn_c3p3/bias:0 | (256,) | -0.0064 | +0.0058 | +0.0021 |
fpn_c2p2/kernel:0 | (1, 1, 256, 256) | -0.0370 | +0.0534 | +0.0064 |
fpn_c2p2/bias:0 | (256,) | -0.0056 | +0.0067 | +0.0021 |
fpn_p5/kernel:0 | (3, 3, 256, 256) | -0.0349 | +0.0396 | +0.0059 |
fpn_p5/bias:0 | (256,) | -0.0089 | +0.0080 | +0.0038 |
fpn_p2/kernel:0 | (3, 3, 256, 256) | -0.0302 | +0.0326 | +0.0055 |
fpn_p2/bias:0 | (256,) | -0.0069 | +0.0060 | +0.0023 |
fpn_p3/kernel:0 | (3, 3, 256, 256) | -0.0248 | +0.0277 | +0.0048 |
fpn_p3/bias:0 | (256,) | -0.0040 | +0.0041 | +0.0015 |
fpn_p4/kernel:0 | (3, 3, 256, 256) | -0.0274 | +0.0281 | +0.0051 |
fpn_p4/bias:0 | (256,) | -0.0041 | +0.0038 | +0.0017 |
rpn_conv_shared/kernel:0 | (3, 3, 256, 512) | -0.0407 | +0.0367 | +0.0027 |
rpn_conv_shared/bias:0 | (512,) | -0.0049 | +0.0074 | +0.0012 |
rpn_class_raw/kernel:0 | (1, 1, 512, 6) | -0.1370 | +0.1370 | +0.0170 |
rpn_class_raw/bias:0 | (6,) | -0.0246 | +0.0246 | +0.0162 |
rpn_bbox_pred/kernel:0 | (1, 1, 512, 12) | -0.1086 | +0.2623 | +0.0242 |
rpn_bbox_pred/bias:0 | (12,) | -0.0504 | +0.0705 | +0.0348 |
mrcnn_class_conv1/kernel:0 | (7, 7, 256, 1024) | -0.0270 | +0.0269 | +0.0033 |
mrcnn_class_conv1/bias:0 | (1024,) | -0.0016 | +0.0005 | +0.0003 |
mrcnn_class_bn1/gamma:0 | (1024,) | +0.9535 | +1.0465 | +0.0092 |
mrcnn_class_bn1/beta:0 | (1024,) | -0.0325 | +0.0055 | +0.0036 |
mrcnn_class_bn1/moving_mean:0 | (1024,) | -13.7297 | +4.7460 | +1.4822 |
mrcnn_class_bn1/moving_variance:0 | (1024,) | +1.6954 | +30.5685 | +2.7805 |
mrcnn_class_conv2/kernel:0 | (1, 1, 1024, 1024) | -0.0782 | +0.0445 | +0.0059 |
mrcnn_class_conv2/bias:0 | (1024,) | -0.0212 | +0.0286 | +0.0053 |
mrcnn_class_bn2/gamma:0 | (1024,) | +0.9765 | +1.0521 | +0.0101 |
mrcnn_class_bn2/beta:0 | (1024,) | -0.0147 | +0.0292 | +0.0045 |
mrcnn_class_bn2/moving_mean:0 | (1024,) | -0.7209 | +0.9867 | +0.1617 |
mrcnn_class_bn2/moving_variance:0 | (1024,) | +0.0094 | +0.3570 | +0.0415 |
mrcnn_class_logits/kernel:0 | (1024, 2) | -0.0778 | +0.0790 | +0.0439 |
mrcnn_class_logits/bias:0 | (2,) | -0.0004 | +0.0004 | +0.0004 |
mrcnn_bbox_fc/kernel:0 | (1024, 8) | -0.0771 | +0.0770 | +0.0432 |
mrcnn_bbox_fc/bias:0 | (8,) | -0.0006 | +0.0004 | +0.0003 |
mrcnn_mask_conv1/kernel:0 | (3, 3, 256, 256) | -0.0690 | +0.0600 | +0.0050 |
mrcnn_mask_conv1/bias:0 | (256,) | -0.0033 | +0.0034 | +0.0010 |
mrcnn_mask_bn1/gamma:0 | (256,) | +0.9733 | +1.1360 | +0.0178 |
mrcnn_mask_bn1/beta:0 | (256,) | -0.0205 | +0.0023 | +0.0036 |
mrcnn_mask_bn1/moving_mean:0 | (256,) | -2.4143 | +1.1433 | +0.5678 |
mrcnn_mask_bn1/moving_variance:0 | (256,) | +0.6905 | +4.3261 | +0.5650 |
mrcnn_mask_conv2/kernel:0 | (3, 3, 256, 256) | -0.0757 | +0.1481 | +0.0050 |
mrcnn_mask_conv2/bias:0 | (256,) | -0.0069 | +0.0069 | +0.0024 |
mrcnn_mask_bn2/gamma:0 | (256,) | +0.9765 | +1.0491 | +0.0119 |
mrcnn_mask_bn2/beta:0 | (256,) | -0.0182 | +0.0021 | +0.0035 |
mrcnn_mask_bn2/moving_mean:0 | (256,) | -0.6003 | +0.1184 | +0.1236 |
mrcnn_mask_bn2/moving_variance:0 | (256,) | +0.0437 | +0.4538 | +0.0551 |
mrcnn_mask_conv3/kernel:0 | (3, 3, 256, 256) | -0.0550 | +0.0605 | +0.0048 |
mrcnn_mask_conv3/bias:0 | (256,) | -0.0118 | +0.0102 | +0.0038 |
mrcnn_mask_bn3/gamma:0 | (256,) | +0.9797 | +1.0433 | +0.0099 |
mrcnn_mask_bn3/beta:0 | (256,) | -0.0307 | +0.0014 | +0.0046 |
mrcnn_mask_bn3/moving_mean:0 | (256,) | -0.5812 | +0.2401 | +0.1452 |
mrcnn_mask_bn3/moving_variance:0 | (256,) | +0.0137 | +0.3976 | +0.0564 |
mrcnn_mask_conv4/kernel:0 | (3, 3, 256, 256) | -0.0425 | +0.0374 | +0.0043 |
mrcnn_mask_conv4/bias:0 | (256,) | -0.0022 | +0.0051 | +0.0010 |
mrcnn_mask_bn4/gamma:0 | (256,) | +0.9905 | +1.0759 | +0.0209 |
mrcnn_mask_bn4/beta:0 | (256,) | +0.0045 | +0.0464 | +0.0113 |
mrcnn_mask_bn4/moving_mean:0 | (256,) | -0.1459 | +0.6716 | +0.1304 |
mrcnn_mask_bn4/moving_variance:0 | (256,) | +0.0331 | +0.2902 | +0.0494 |
mrcnn_mask_deconv/kernel:0 | (2, 2, 256, 256) | -0.0395 | +0.0652 | +0.0057 |
mrcnn_mask_deconv/bias:0 | (256,) | -0.0030 | +0.0770 | +0.0112 |
mrcnn_mask/kernel:0 | (1, 1, 256, 2) | -0.1647 | +0.1596 | +0.0868 |
mrcnn_mask/bias:0 | (2,) | +0.0000 | +0.0048 | +0.0024 |
Prediction on a Random Validation Image
image_id = random.choice(dataset.image_ids)
image, image_meta, gt_class_id, gt_bbox, gt_mask =\
modellib.load_image_gt(dataset, config, image_id, use_mini_mask=False)
info = dataset.image_info[image_id]
print("image ID: {}.{} ({}) {}".format(info["source"], info["id"], image_id,
dataset.image_reference(image_id)))
# Run object detection
results = model.detect([image], verbose=1)
# Display results
ax = get_ax(1)
r = results[0]
visualize.display_instances(image, r['rois'], r['masks'], r['class_ids'],
dataset.class_names, r['scores'], ax=ax,
title="Predictions")
log("gt_class_id", gt_class_id)
log("gt_bbox", gt_bbox)
log("gt_mask", gt_mask)
print('The car has:{} damages'.format(len(dataset.image_info[image_id]['polygons'])))
image ID: scratch.image53.jpeg (2) C:/Users/Sourish/Mask_RCNN/custom/val\image53.jpeg
Processing 1 images
image shape: (1024, 1024, 3) min: 0.00000 max: 255.00000 uint8
molded_images shape: (1, 1024, 1024, 3) min: -123.70000 max: 151.10000 float64
image_metas shape: (1, 14) min: 0.00000 max: 1024.00000 int32
anchors shape: (1, 261888, 4) min: -0.35390 max: 1.29134 float32
gt_class_id shape: (2,) min: 1.00000 max: 1.00000 int32
gt_bbox shape: (2, 4) min: 315.00000 max: 728.00000 int32
gt_mask shape: (1024, 1024, 2) min: 0.00000 max: 1.00000 bool
The car has:2 damages
image_id = random.choice(dataset.image_ids)
image, image_meta, gt_class_id, gt_bbox, gt_mask =\
modellib.load_image_gt(dataset, config, image_id, use_mini_mask=False)
info = dataset.image_info[image_id]
print("image ID: {}.{} ({}) {}".format(info["source"], info["id"], image_id,
dataset.image_reference(image_id)))
# Run object detection
results = model.detect([image], verbose=1)
# Display results
ax = get_ax(1)
r = results[0]
visualize.display_instances(image, r['rois'], r['masks'], r['class_ids'],
dataset.class_names, r['scores'], ax=ax,
title="Predictions")
log("gt_class_id", gt_class_id)
log("gt_bbox", gt_bbox)
log("gt_mask", gt_mask)
print('The car has:{} damages'.format(len(dataset.image_info[image_id]['polygons'])))
image ID: scratch.image52.jpeg (1) C:/Users/Sourish/Mask_RCNN/custom/val\image52.jpeg
Processing 1 images
image shape: (1024, 1024, 3) min: 0.00000 max: 255.00000 uint8
molded_images shape: (1, 1024, 1024, 3) min: -123.70000 max: 141.10000 float64
image_metas shape: (1, 14) min: 0.00000 max: 1024.00000 int32
anchors shape: (1, 261888, 4) min: -0.35390 max: 1.29134 float32
gt_class_id shape: (1,) min: 1.00000 max: 1.00000 int32
gt_bbox shape: (1, 4) min: 272.00000 max: 930.00000 int32
gt_mask shape: (1024, 1024, 1) min: 0.00000 max: 1.00000 bool
The car has:1 damages
Pretty decent prediction considering training with only 49 images and 15 epochs