mask rcnn网络结构笔记
生活随笔
收集整理的這篇文章主要介紹了
mask rcnn网络结构笔记
小編覺得挺不錯的,現在分享給大家,幫大家做個參考.
基于https://gluon-cv.mxnet.io/build/examples_instance/demo_mask_rcnn.html?調試打印得到net.summary(x)
--------------------------------------------------------------------------------Layer (type) Output Shape Param #
================================================================================Input (1, 3, 600, 800) 0Conv2D-1 (1, 64, 300, 400) 9408BatchNorm-2 (1, 64, 300, 400) 256Activation-3 (1, 64, 300, 400) 0MaxPool2D-4 (1, 64, 150, 200) 0Conv2D-5 (1, 64, 150, 200) 4096BatchNorm-6 (1, 64, 150, 200) 256Activation-7 (1, 64, 150, 200) 0Conv2D-8 (1, 64, 150, 200) 36864BatchNorm-9 (1, 64, 150, 200) 256Activation-10 (1, 64, 150, 200) 0Conv2D-11 (1, 256, 150, 200) 16384BatchNorm-12 (1, 256, 150, 200) 1024Conv2D-13 (1, 256, 150, 200) 16384BatchNorm-14 (1, 256, 150, 200) 1024Activation-15 (1, 256, 150, 200) 0BottleneckV1b-16 (1, 256, 150, 200) 0Conv2D-17 (1, 64, 150, 200) 16384BatchNorm-18 (1, 64, 150, 200) 256Activation-19 (1, 64, 150, 200) 0Conv2D-20 (1, 64, 150, 200) 36864BatchNorm-21 (1, 64, 150, 200) 256Activation-22 (1, 64, 150, 200) 0Conv2D-23 (1, 256, 150, 200) 16384BatchNorm-24 (1, 256, 150, 200) 1024Activation-25 (1, 256, 150, 200) 0BottleneckV1b-26 (1, 256, 150, 200) 0Conv2D-27 (1, 64, 150, 200) 16384BatchNorm-28 (1, 64, 150, 200) 256Activation-29 (1, 64, 150, 200) 0Conv2D-30 (1, 64, 150, 200) 36864BatchNorm-31 (1, 64, 150, 200) 256Activation-32 (1, 64, 150, 200) 0Conv2D-33 (1, 256, 150, 200) 16384BatchNorm-34 (1, 256, 150, 200) 1024Activation-35 (1, 256, 150, 200) 0BottleneckV1b-36 (1, 256, 150, 200) 0Conv2D-37 (1, 128, 150, 200) 32768BatchNorm-38 (1, 128, 150, 200) 512Activation-39 (1, 128, 150, 200) 0Conv2D-40 (1, 128, 75, 100) 147456BatchNorm-41 (1, 128, 75, 100) 512Activation-42 (1, 128, 75, 100) 0Conv2D-43 (1, 512, 75, 100) 65536BatchNorm-44 (1, 512, 75, 100) 2048Conv2D-45 (1, 512, 75, 100) 131072BatchNorm-46 (1, 512, 75, 100) 2048Activation-47 (1, 512, 75, 100) 0BottleneckV1b-48 (1, 512, 75, 100) 0Conv2D-49 (1, 128, 75, 100) 65536BatchNorm-50 (1, 128, 75, 100) 512Activation-51 (1, 128, 75, 100) 0Conv2D-52 (1, 128, 75, 100) 147456BatchNorm-53 (1, 128, 75, 100) 512Activation-54 (1, 128, 75, 100) 0Conv2D-55 (1, 512, 75, 100) 65536BatchNorm-56 (1, 512, 75, 100) 2048Activation-57 (1, 512, 75, 100) 0BottleneckV1b-58 (1, 512, 75, 100) 0Conv2D-59 (1, 128, 75, 100) 65536BatchNorm-60 (1, 128, 75, 100) 512Activation-61 (1, 128, 75, 100) 0Conv2D-62 (1, 128, 75, 100) 147456BatchNorm-63 (1, 128, 75, 100) 512Activation-64 (1, 128, 75, 100) 0Conv2D-65 (1, 512, 75, 100) 65536BatchNorm-66 (1, 512, 75, 100) 2048Activation-67 (1, 512, 75, 100) 0BottleneckV1b-68 (1, 512, 75, 100) 0Conv2D-69 (1, 128, 75, 100) 65536BatchNorm-70 (1, 128, 75, 100) 512Activation-71 (1, 128, 75, 100) 0Conv2D-72 (1, 128, 75, 100) 147456BatchNorm-73 (1, 128, 75, 100) 512Activation-74 (1, 128, 75, 100) 0Conv2D-75 (1, 512, 75, 100) 65536BatchNorm-76 (1, 512, 75, 100) 2048Activation-77 (1, 512, 75, 100) 0BottleneckV1b-78 (1, 512, 75, 100) 0Conv2D-79 (1, 256, 75, 100) 131072BatchNorm-80 (1, 256, 75, 100) 1024Activation-81 (1, 256, 75, 100) 0Conv2D-82 (1, 256, 38, 50) 589824BatchNorm-83 (1, 256, 38, 50) 1024Activation-84 (1, 256, 38, 50) 0Conv2D-85 (1, 1024, 38, 50) 262144BatchNorm-86 (1, 1024, 38, 50) 4096Conv2D-87 (1, 1024, 38, 50) 524288BatchNorm-88 (1, 1024, 38, 50) 4096Activation-89 (1, 1024, 38, 50) 0BottleneckV1b-90 (1, 1024, 38, 50) 0Conv2D-91 (1, 256, 38, 50) 262144BatchNorm-92 (1, 256, 38, 50) 1024Activation-93 (1, 256, 38, 50) 0Conv2D-94 (1, 256, 38, 50) 589824BatchNorm-95 (1, 256, 38, 50) 1024Activation-96 (1, 256, 38, 50) 0Conv2D-97 (1, 1024, 38, 50) 262144BatchNorm-98 (1, 1024, 38, 50) 4096Activation-99 (1, 1024, 38, 50) 0BottleneckV1b-100 (1, 1024, 38, 50) 0Conv2D-101 (1, 256, 38, 50) 262144BatchNorm-102 (1, 256, 38, 50) 1024Activation-103 (1, 256, 38, 50) 0Conv2D-104 (1, 256, 38, 50) 589824BatchNorm-105 (1, 256, 38, 50) 1024Activation-106 (1, 256, 38, 50) 0Conv2D-107 (1, 1024, 38, 50) 262144BatchNorm-108 (1, 1024, 38, 50) 4096Activation-109 (1, 1024, 38, 50) 0BottleneckV1b-110 (1, 1024, 38, 50) 0Conv2D-111 (1, 256, 38, 50) 262144BatchNorm-112 (1, 256, 38, 50) 1024Activation-113 (1, 256, 38, 50) 0Conv2D-114 (1, 256, 38, 50) 589824BatchNorm-115 (1, 256, 38, 50) 1024Activation-116 (1, 256, 38, 50) 0Conv2D-117 (1, 1024, 38, 50) 262144BatchNorm-118 (1, 1024, 38, 50) 4096Activation-119 (1, 1024, 38, 50) 0BottleneckV1b-120 (1, 1024, 38, 50) 0Conv2D-121 (1, 256, 38, 50) 262144BatchNorm-122 (1, 256, 38, 50) 1024Activation-123 (1, 256, 38, 50) 0Conv2D-124 (1, 256, 38, 50) 589824BatchNorm-125 (1, 256, 38, 50) 1024Activation-126 (1, 256, 38, 50) 0Conv2D-127 (1, 1024, 38, 50) 262144BatchNorm-128 (1, 1024, 38, 50) 4096Activation-129 (1, 1024, 38, 50) 0BottleneckV1b-130 (1, 1024, 38, 50) 0Conv2D-131 (1, 256, 38, 50) 262144BatchNorm-132 (1, 256, 38, 50) 1024Activation-133 (1, 256, 38, 50) 0Conv2D-134 (1, 256, 38, 50) 589824BatchNorm-135 (1, 256, 38, 50) 1024Activation-136 (1, 256, 38, 50) 0Conv2D-137 (1, 1024, 38, 50) 262144BatchNorm-138 (1, 1024, 38, 50) 4096Activation-139 (1, 1024, 38, 50) 0BottleneckV1b-140 (1, 1024, 38, 50) 0
FasterRCNN.features
---------------------------------------------------------------------------------------------
RPNAnchorGenerator-141 (1, 28500, 4) 983040
RPN.anchor_generator
---------------------------------------------------------------------------------------------Conv2D-142 (1, 1024, 38, 50) 9438208Activation-143 (1, 1024, 38, 50) 0
RPN.conv1
---------------------------------------------------------------------------------------------Conv2D-144 (1, 15, 38, 50) 15375
RPN.score
---------------------------------------------------------------------------------------------Conv2D-145 (1, 60, 38, 50) 61500
RPN.loc
---------------------------------------------------------------------------------------------
BBoxCornerToCenter-146 (1, 28500, 4) 0
RPNProposal._box_to_center
---------------------------------------------------------------------------------------------
NormalizedBoxCenterDecoder-147 (1, 28500, 4) 0
RPNProposal._box_decoder
---------------------------------------------------------------------------------------------BBoxClipToImage-148 (1, 28500, 4) 0
RPNProposal._clipper
---------------------------------------------------------------------------------------------RPNProposal-149 (1, 28500, 5) 0
RPN.region_proposer
---------------------------------------------------------------------------------------------RPN-150 (1, 1000, 1), (1, 1000, 4) 0
FasterRCNN.rpn
---------------------------------------------------------------------------------------------Conv2D-151 (1000, 512, 14, 14) 524288BatchNorm-152 (1000, 512, 14, 14) 2048Activation-153 (1000, 512, 14, 14) 0Conv2D-154 (1000, 512, 7, 7) 2359296BatchNorm-155 (1000, 512, 7, 7) 2048Activation-156 (1000, 512, 7, 7) 0Conv2D-157 (1000, 2048, 7, 7) 1048576BatchNorm-158 (1000, 2048, 7, 7) 8192Conv2D-159 (1000, 2048, 7, 7) 2097152BatchNorm-160 (1000, 2048, 7, 7) 8192Activation-161 (1000, 2048, 7, 7) 0BottleneckV1b-162 (1000, 2048, 7, 7) 0Conv2D-163 (1000, 512, 7, 7) 1048576BatchNorm-164 (1000, 512, 7, 7) 2048Activation-165 (1000, 512, 7, 7) 0Conv2D-166 (1000, 512, 7, 7) 2359296BatchNorm-167 (1000, 512, 7, 7) 2048Activation-168 (1000, 512, 7, 7) 0Conv2D-169 (1000, 2048, 7, 7) 1048576BatchNorm-170 (1000, 2048, 7, 7) 8192Activation-171 (1000, 2048, 7, 7) 0BottleneckV1b-172 (1000, 2048, 7, 7) 0Conv2D-173 (1000, 512, 7, 7) 1048576BatchNorm-174 (1000, 512, 7, 7) 2048Activation-175 (1000, 512, 7, 7) 0Conv2D-176 (1000, 512, 7, 7) 2359296BatchNorm-177 (1000, 512, 7, 7) 2048Activation-178 (1000, 512, 7, 7) 0Conv2D-179 (1000, 2048, 7, 7) 1048576BatchNorm-180 (1000, 2048, 7, 7) 8192Activation-181 (1000, 2048, 7, 7) 0BottleneckV1b-182 (1000, 2048, 7, 7) 0
FasterRCNN.top_features
---------------------------------------------------------------------------------------------Dense-183 (1000, 81) 165969
FasterRCNN.class_predictor
---------------------------------------------------------------------------------------------Dense-184 (1000, 320) 655680
FasterRCNN.box_predictor
---------------------------------------------------------------------------------------------
MultiPerClassDecoder-185 (1, 1000, 80), (1, 1000, 80) 0
FasterRCNN.cls_decoder
---------------------------------------------------------------------------------------------
BBoxCornerToCenter-186 (1, 1000, 4) 0
FasterRCNN.box_to_center
---------------------------------------------------------------------------------------------
NormalizedBoxCenterDecoder-187 (80, 1000, 4) 0
FasterRCNN.box_decoder
---------------------------------------------------------------------------------------------Conv2D-188 (1000, 512, 14, 14) 524288BatchNorm-189 (1000, 512, 14, 14) 2048Activation-190 (1000, 512, 14, 14) 0Conv2D-191 (1000, 512, 7, 7) 2359296BatchNorm-192 (1000, 512, 7, 7) 2048Activation-193 (1000, 512, 7, 7) 0Conv2D-194 (1000, 2048, 7, 7) 1048576BatchNorm-195 (1000, 2048, 7, 7) 8192Conv2D-196 (1000, 2048, 7, 7) 2097152BatchNorm-197 (1000, 2048, 7, 7) 8192Activation-198 (1000, 2048, 7, 7) 0BottleneckV1b-199 (1000, 2048, 7, 7) 0Conv2D-200 (1000, 512, 7, 7) 1048576BatchNorm-201 (1000, 512, 7, 7) 2048Activation-202 (1000, 512, 7, 7) 0Conv2D-203 (1000, 512, 7, 7) 2359296BatchNorm-204 (1000, 512, 7, 7) 2048Activation-205 (1000, 512, 7, 7) 0Conv2D-206 (1000, 2048, 7, 7) 1048576BatchNorm-207 (1000, 2048, 7, 7) 8192Activation-208 (1000, 2048, 7, 7) 0BottleneckV1b-209 (1000, 2048, 7, 7) 0Conv2D-210 (1000, 512, 7, 7) 1048576BatchNorm-211 (1000, 512, 7, 7) 2048Activation-212 (1000, 512, 7, 7) 0Conv2D-213 (1000, 512, 7, 7) 2359296BatchNorm-214 (1000, 512, 7, 7) 2048Activation-215 (1000, 512, 7, 7) 0Conv2D-216 (1000, 2048, 7, 7) 1048576BatchNorm-217 (1000, 2048, 7, 7) 8192Activation-218 (1000, 2048, 7, 7) 0BottleneckV1b-219 (1000, 2048, 7, 7) 0
MaskRCNN.top_features
---------------------------------------------------------------------------------------------Conv2DTranspose-220 (1000, 256, 14, 14) 2097408Conv2D-221 (1000, 80, 14, 14) 20560Mask-222 (1, 1000, 80, 14, 14) 0
MaskRCNN.mask
---------------------------------------------------------------------------------------------MaskRCNN-223 (1, 1000, 1), (1, 1000, 1), (1, 1000, 4), (1, 1000, 14, 14) 0
================================================================================
Parameters in forward computation graph, duplicate includedTotal params: 51986156Trainable params: 50927468Non-trainable params: 1058688
Shared params in forward computation graph: 14987264
Unique parameters in model: 36998892
--------------------------------------------------------------------------------
?
總結
以上是生活随笔為你收集整理的mask rcnn网络结构笔记的全部內容，希望文章能夠幫你解決所遇到的問題。
- 上一篇： yolov3网络结构笔记
- 下一篇： 理解YOLOv2训练过程中输出参数含义