From bcc6334ed3a2aac2cc4e98f0d63fd77d4c9df798 Mon Sep 17 00:00:00 2001 From: wassname Date: Sun, 2 Oct 2016 10:05:52 +0800 Subject: [PATCH] update readme --- images/dice_equation.png | Bin 0 -> 2187 bytes readme.md | 13 ++++++------- 2 files changed, 6 insertions(+), 7 deletions(-) create mode 100644 images/dice_equation.png diff --git a/images/dice_equation.png b/images/dice_equation.png new file mode 100644 index 0000000000000000000000000000000000000000..ed691bce52a387ce56e7d070e35775ef930893ab GIT binary patch literal 2187 zcmZ`*c{tQ-8~%~9G}%(f@{Jn1(HMq|K^TTC8Jr@6$*yRMv5cjLV;vb&7;9$=lY?+j zofumx`w~eiTOk^zWh_IaGj+ayzCXVAx}NvGp8I~D=e@7@pZCphcETJGJ0b=EzyUj3 zv_wStF%g+TOTrieE?U3R}eggKhb#VlMSTz73 zCj-DYK1E&vfJhhsu<-zZEC2x6%O%ZN6kibZwa1`=A8FYG?fgvivaLr1A8mhyP47iB z0NA%~hqiEwdBZ8;+$gq{Ol35M1Ab{_{f;(}^S*H7BtL!H^3g|~^0T1YN1*36m%9Ty zQt`PKGZJQ?r6Vibf@^XY^5c|~N)%Fl z^U1k6oT$hk+BQlFJx$I8S!Ac;Qd#z5AF zMzg%#q(0$Y#3d5AV((1XjYp^Scw$@n%{%c)Z~o5gxj_h4Rx)h+LaK^8#LY(uv^FV< zoG|J?b#^tZH(%v6>IQvqd)({GzFRxWDh+0KMP6*%ex+MF)5dA8V-duPy!-5%BWRgH z0|S=dD@g9#@5Or>1mAxW-7FB z&C1D{;JdjdE>Ty1Hf;#8q##Ne@5!1PZTiqAumdt$LvHFcK%*Z_i~^5FEW&eu`4=s>zv&O>-K)Yo`cs^DU5@>qju zQNf$!(JDRNz644H7oxo=Ws_(+YRd9dr2AU;`F5nKO3e%-miF#hw;}9^r8?d%8{Xd9A7xZ~3#v zXO$&AtMP|jxlFlLxGYY={u&Re1)<$4J4qZ&)^e$*9KO8OfpxlGZ(uVQa@(=&xBb;Y z+y;)eR#rl;O4p%*UzOv{(&o|K`mhr7GhW@l~ z-L*q7KCXH}XQQ8(2p^(t97ZBpg9;RG3_7Rk!|xJZ)VB+386~-x4q2wBLXo~^_+-P8{{?pmcG0hAwFpnnQIhCPxkK@~J-9y?GON#&$oXv>lXADSO`s5OwPKHeT> zJF(nk-n4U*UVxn*npf$qsZyKSGQ+N#bzV_zIyqg~+fG@3(Edqk`$(yHypUN$Nxf)8 z#=^y1PyHs!)^P0x(q@GX3!Y+s-x;AGJ4ipU0a{)Sm38OPh_Ix8_h#^bM$@j^OE%ts^jN@H5 zIaThr-Ec`w>}taHSj_wD&aqost~$j!c|~=cx~}XI?KqR!N$F2BkgyQ+#xbYOsqStZ zZO%x+dAVl}(!c5Mvgin&`Br3)4R+r@BKWZWw9bTt>)Ofb6ETj#hKcg@#*M4*2H#DH zta-@nec~c~FeSZ!Om6)?k#bN@{a=IJaWKOg3)Ar^)X1;@vDKjOj-#1uYQ2COd1w(p zT#`&mQ6API_dG3op&HaU8Ccn3+_UH`b85e-L$1w7Z(X|@Z>P4tn0`ZgpJ#GB-1?4S z#m4JS^c8j#hIHH~pd~=Tej#>ByM0nkUsFsAMrSbO*U|*`K!Q2=ZzX|$U^a1YkFE}w z>%{2Os&mVBtAFC!{UC$D<4nKkvNE8px=Q0XJ$loG4B zZj=yx!k&5N^LBfnt4wU6#vV3)&)xhw_{)YYW_arvr2g7XbPB3Ry?eD_z9kHmw)U!` z+W*bttX*W*!}w(poN1@*Z4J!*+u_~unRBnA~{O-JZJ!s*wZaJ%F z>W2VGbWK&7b=8jl5~>od-HHC^i2+Ez@BlslaG0JUR1XG)!QEhbNCX0@Z=el>Az`pL zL880fq)&rreHlNE-?JOUmO??#zjPi`Gmw< z_=JEh`KS%X`3Hm&&nLwFD}eceHyo_}Q~ZVhl_8LG;5KpoaL~`;euPj$H~{;#IvE>x k_`LW}K0F|T5E@AgAzb2{gsn$M{`@TfcGgblT1$M=KZPpw`2YX_ literal 0 HcmV?d00001 diff --git a/readme.md b/readme.md index dcac4d4..f0d8c57 100644 --- a/readme.md +++ b/readme.md @@ -24,7 +24,6 @@ For data augmentation, we used a range of flips, affine transforms, and jitter. ### Model - ![model_diagram](images/model_diagram.png) *The model architecture. Every box is an inception module or convolution with the number of feature layers denoted in brackets. The output size is listed below each box and the coloured arrows denote different operations.* @@ -52,13 +51,13 @@ The final results gave an accuracy of 0.73 for the test data set and 0.745 for t The image below shows results and displays the weaknesses of this model. Firstly, it fails where foliage obscures large parts of the pipe, and secondly, it occasionally returns false positives for bleached wood or similar objects. This may be overcome with further training data or by using a region proposal network to evaluate pipe shaped regions. -results|training|testing --|-|- -**accuracy**|0.745|0.73 -**dice loss**|0.079|0.100 -**mathews correlation coeffecient**|0.82|0.77 -*Results on unaugmented training and test data* +|results|training|testing| +|-|-|-| +|**accuracy**|0.745|0.73| +|**dice loss**|0.079|0.100| +|**mathews correlation coeffecient**|0.82|0.77| +*Results on unaugmented training and test data* ![results](images/results4.png) *Results for test data. Columns: (a) the input data, (b) the ground truth mask (c) the mask predicted by this model.*