# HG changeset patch
# User Laman
# Date 2019-06-09 21:10:09
# Node ID db53fefbf5575749d855df0bd59ded908917c029
# Parent  d4478db55eec9976c9dc3b2942524b3d681d4235

Hakugen detection module

diff --git a/exp/dochi.py b/exp/dochi.py
--- a/exp/dochi.py
+++ b/exp/dochi.py
@@ -3,12 +3,12 @@ import time
 import argparse
 import logging as log
 
+import cv2 as cv
 import numpy as np
 import keras
 from keras.models import load_model
-from PIL import Image,ImageDraw
+from PIL import Image
 
-from epoint import EPoint
 import exp_config as cfg
 from kerokero.k_util import averageDistance
 
@@ -16,6 +16,7 @@ keras.losses.averageDistance=averageDist
 keras.metrics.averageDistance=averageDistance
 
 model=load_model(cfg.dochiModel)
+SIDE=224
 
 
 def locateGrid(img):
@@ -28,7 +29,7 @@ def locateGrid(img):
 	label=model.predict(np.reshape(npImg,(1,224,224,1)))
 	points=[]
 	for i in range(4):
-		points.append(EPoint((label[0][i][0]+1)*(width/2),(label[0][i][1]+1)*(height/2)))
+		points.append([(label[0][i][0]+1)*(width/2),(label[0][i][1]+1)*(height/2)])
 	t=time.time()-t1
 	log.info("grid located in {0:.3}s".format(t))
 	return points
@@ -43,7 +44,10 @@ if __name__=="__main__":
 	for image_path in args.input:
 		image=Image.open(image_path)
 		points=locateGrid(image)
-		drawer=ImageDraw.Draw(image)
-		for p in points:
-			drawer.ellipse((p.x-2,p.y-2,p.x+2,p.y+2),fill="#00ff00")
-		image.save(os.path.join(args.output_dir,os.path.basename(image_path)))
+		x1=SIDE*0.1
+		x2=SIDE*0.9
+		destPoints=[(x1,x1),(x1,x2),(x2,x2),(x2,x1)]
+		m=cv.getPerspectiveTransform(np.float32(points),np.float32(destPoints))
+		img=cv.warpPerspective(np.uint8(image),m,(SIDE,SIDE))
+		img=cv.cvtColor(img,cv.COLOR_BGR2RGB)
+		cv.imwrite(os.path.join(args.output_dir,os.path.basename(image_path)),img)
diff --git a/exp/exp_config.py b/exp/exp_config.py
--- a/exp/exp_config.py
+++ b/exp/exp_config.py
@@ -13,6 +13,7 @@ INTERACTIVE=False
 imgDir="/tmp/oneEye"
 sansaModel=cfg.get("sansaModel")
 dochiModel=cfg.get("dochiModel")
+hakugenModel=cfg.get("hakugenModel")
 i=1
 
 if not os.path.exists(imgDir):
diff --git a/exp/hakugen.py b/exp/hakugen.py
new file mode 100644
--- /dev/null
+++ b/exp/hakugen.py
@@ -0,0 +1,49 @@
+import os
+import time
+import argparse
+import logging as log
+
+import numpy as np
+import keras
+from keras.models import load_model
+from PIL import Image,ImageDraw
+
+from epoint import EPoint
+import exp_config as cfg
+from kerokero.k_util import averageDistance
+
+keras.losses.averageDistance=averageDistance
+keras.metrics.averageDistance=averageDistance
+
+model=load_model(cfg.hakugenModel)
+
+
+def locateGrid(img):
+	t1=time.time()
+	(width,height)=img.size
+	normedImg=img.convert("L")
+	npImg=np.array(normedImg.getdata()).reshape((224,224,1)).astype(np.float32)
+	npImg=npImg/128-1
+
+	label=model.predict(np.reshape(npImg,(1,224,224,1)))
+	points=[]
+	for i in range(4):
+		points.append(EPoint((label[0][i][0]+1)*(width/2),(label[0][i][1]+1)*(height/2)))
+	t=time.time()-t1
+	log.info("grid located in {0:.3}s".format(t))
+	return points
+
+
+if __name__=="__main__":
+	parser=argparse.ArgumentParser()
+	parser.add_argument("-i","--input",nargs="+")
+	parser.add_argument("-o","--output_dir",required=True)
+	args=parser.parse_args()
+
+	for image_path in args.input:
+		image=Image.open(image_path)
+		points=locateGrid(image)
+		drawer=ImageDraw.Draw(image)
+		for p in points:
+			drawer.ellipse((p.x-2,p.y-2,p.x+2,p.y+2),fill="#00ff00")
+		image.save(os.path.join(args.output_dir,os.path.basename(image_path)))
diff --git a/exp/kerokero/prepare_data.py b/exp/kerokero/prepare_data.py
--- a/exp/kerokero/prepare_data.py
+++ b/exp/kerokero/prepare_data.py
@@ -1,6 +1,7 @@
 import os
 import sys
 import re
+import math
 import random
 import logging as log
 
@@ -48,6 +49,19 @@ class Sample:
 		grid=list(map(lambda p: list(2*p/self.SIDE-EPoint(1,1)), grid))
 		return (img,grid)
 
+	def rectify(self):
+		x1=self.SIDE*0.1
+		x2=self.SIDE*0.9
+		abcd=list(map(list,self.grid))
+		destPoints=[(x1,x1),(x1,x2),(x2,x2),(x2,x1)]
+		abcd_=list(map(list, (EPoint(x,y)+self._createNoise() for (x,y) in destPoints)))
+		m=cv.getPerspectiveTransform(np.float32(abcd),np.float32(abcd_))
+		img=cv.warpPerspective(self.img,m,(self.SIDE,self.SIDE))
+		img=np.uint8(img)
+		grid=Corners(c.transform(m) for c in self.grid)
+		grid=list(map(lambda p: list(2*p/self.SIDE-EPoint(1,1)), grid))
+		return (img,grid)
+
 	def _getCenter(self):
 		(a,b,c,d)=self.grid
 		p=Line.fromPoints(a,c)
@@ -63,6 +77,11 @@ class Sample:
 		scale=getScale(self.SIDE/(wg*(1+left+right)), self.SIDE/(hg*(1+top+bottom)))
 		return np.matmul(scale,t2)
 
+	def _createNoise(self):
+		alpha=random.uniform(0,math.pi*2)
+		d=random.uniform(0,self.SIDE*0.05)
+		return EPoint(math.cos(alpha)*d, math.sin(alpha)*d)
+
 	def show(self):
 		img=cv.cvtColor(self.img,cv.COLOR_GRAY2BGR)
 		for c in self.grid:
@@ -91,14 +110,15 @@ def harvestDir(path):
 	for f in files:
 		grade=annotations.get(f.name,[Board()])[0].grade
 		Stats.counts[grade]+=1
-		if not Board.UNSET<grade<=Board.POOR: continue
+		if not Board.UNSET<grade<=Board.GOOD: continue
 		img=cv.imread(f.path)
 		img=cv.cvtColor(img,cv.COLOR_BGR2GRAY)
 		for b in boards:
 			sample=Sample(img,b.grid)
 			# sample.show()
-			(transformedImg,label)=sample.transform()
-			# Sample(np.uint8((transformedImg+1)*128),map(lambda c: (c+EPoint(1,1))*Sample.SIDE/2,transformedGrid)).show()
+			# (transformedImg,label)=sample.transform()
+			(transformedImg,label)=sample.rectify()
+			# Sample(np.uint8(transformedImg),map(lambda c: (EPoint(*c)+EPoint(1,1))*Sample.SIDE/2,label)).show()
 			yield (transformedImg,label)
 
 
diff --git a/exp/kerokero/train.py b/exp/kerokero/train.py
--- a/exp/kerokero/train.py
+++ b/exp/kerokero/train.py
@@ -5,8 +5,8 @@ import argparse
 import logging as log
 
 import numpy as np
-from keras.layers import Conv2D,Dropout,Dense,Flatten,MaxPooling2D,BatchNormalization,GlobalAveragePooling2D,Reshape,concatenate
-from keras.models import Sequential,load_model,Model,Input
+from keras.layers import Conv2D,Dropout,Dense,Flatten,MaxPooling2D,BatchNormalization,GlobalAveragePooling2D,Reshape
+from keras.models import Sequential,load_model
 from keras.callbacks import TensorBoard,ModelCheckpoint
 import keras.metrics
 
@@ -73,41 +73,10 @@ def createCNN():
 	return model
 
 
-def createHinted():
-	input=Input((224,224,1))
-	base=load_model(args.load_hints)
-	for layer in base.layers:
-		layer.trainable=False
-	hints=base(input)
-
-	x=BatchNormalization()(input)
-	x=Conv2D(24,(5,5),padding="same",kernel_initializer="he_normal",activation="relu",input_shape=(224,224,1),data_format="channels_last")(x)
-	x=MaxPooling2D(pool_size=(2,2),strides=(2,2),padding="valid")(x)
-	x=Conv2D(36,(5,5),activation="relu")(x)
-	x=MaxPooling2D(pool_size=(2,2),strides=(2,2),padding="valid")(x)
-	x=Conv2D(48,(5,5),activation="relu")(x)
-	x=MaxPooling2D(pool_size=(2,2),strides=(2,2),padding="valid")(x)
-	x=Conv2D(64,(3,3),activation="relu")(x)
-	x=MaxPooling2D(pool_size=(2,2),strides=(2,2),padding="valid")(x)
-	x=Conv2D(64,(3,3),activation="relu")(x)
-	x=GlobalAveragePooling2D()(x)
-
-	x=concatenate([x,Flatten()(hints)])
-	x=Dense(500,activation="relu")(x)
-	x=Dense(90,activation="relu")(x)
-
-	predictions=Reshape((4,2))(Dense(8)(x))
-
-	model=Model(inputs=input,outputs=predictions)
-
-	model.compile(optimizer='rmsprop',loss=averageDistance,metrics=['mae','accuracy'])
-	return model
-
-
 if args.load_model:
 	model=load_model(args.load_model)
 else:
-	model=createHinted()
+	model=createCNN()
 	model.summary()
 
 log.info("loading data...")