def query_blob(id):
blob = Image("static\\img\\%d.jpg" % id).crop(
int(request.form["x"]), int(request.form["y"]), int(request.form["width"]), int(request.form["height"])
)
if blob and "SK_MODEL" in app.config:
if blob.height > blob.width:
blob = blob.resize(h=app.config["PATCH_SIZE"])
else:
blob = blob.resize(w=app.config["PATCH_SIZE"])
blob = blob.embiggen((app.config["PATCH_SIZE"], app.config["PATCH_SIZE"]))
np_img = blob.getGrayNumpy().transpose().reshape(-1)
pred = labels.inverse_transform(sk_model.predict(np_img))[0]
return jsonify(prediction=pred)
else:
return jsonify(prediction="")
def parse_frame(img):
"""
Parses a SimpleCV image object of a frame from Super Hexagon.
Returns a ParsedFrame object containing selected features.
"""
img
# helper image size variables
w,h = img.size()
midx,midy = w/2,h/2
# Create normalized images for targeting objects in the foreground or background.
# (This normalization is handy since Super Hexagon's colors are inverted for some parts of the game)
# fg_img = foreground image (bright walls, black when binarized)
# bg_img = background image (bright space, black when binarized)
fg_img = img
if sum(img.binarize().getPixel(midx,midy)) == 0:
fg_img = img.invert()
bg_img = fg_img.invert()
# We need to close any gaps around the center wall so we can detect its containing blob.
# The gaps are resulting artifacts from video encoding.
# The 'erode' function does this by expanding the dark parts of the image.
bimg = bg_img.binarize()
bimg = black_out_GUI(bimg)
blobs = bimg.findBlobs()
cursor_blob = get_cursor_blob(blobs, h, midx, midy)
if cursor_blob:
cursor_point = map(int, cursor_blob.centroid())
cursor_r, cursor_angle = cart_to_polar(cursor_point[0] - midx, midy - cursor_point[1])
cursor_angle = int(cursor_angle * 360/ (2 * pi))
cursor_angle = 180 - cursor_angle
if cursor_angle < 0:
a += 360
bimg = black_out_center(bimg, cursor_r).applyLayers()
arr = bimg.resize(100).getGrayNumpy() > 100
rot_arr = arr_to_polar(arr)
rot_img = Image(PIL.Image.fromarray(np.uint8(np.transpose(rot_arr)*255))).dilate(iterations=3)
rot_arr = rot_img.getGrayNumpy() > 100
rot_img = rot_img.resize(400).flipVertical()
return ParsedFrame(img, bimg, arr, rot_arr, rot_img, cursor_r, cursor_angle)
else:
return None
def query_blob(id):
blob = Image("static\\img\\%d.jpg" % id).crop(int(request.form['x']),
int(request.form['y']),
int(request.form['width']),
int(request.form['height']))
if blob and 'SK_MODEL' in app.config:
if blob.height > blob.width:
blob = blob.resize(h=app.config['PATCH_SIZE'])
else:
blob = blob.resize(w=app.config['PATCH_SIZE'])
blob = blob.embiggen(
(app.config['PATCH_SIZE'], app.config['PATCH_SIZE']))
np_img = blob.getGrayNumpy().transpose().reshape(-1)
pred = labels.inverse_transform(sk_model.predict(np_img))[0]
return jsonify(prediction=pred)
else:
return jsonify(prediction='')
def getRotateBoxTest(self):
if not self._rotBoxTest:
grayImg = self.image.toGray()
diff = grayImg - self._boxFilter(grayImg, 35)
result = [0]*18
if not self.ellipse.error and self.ellipse.a > 0 and self.ellipse.b > 0:
res = countInWindowEst(diff.threshold(6).getGrayNumpy(), 5)
res = self._getRotatedAndScaledGalImg(Image(res))
if res != None:
res = res.getGrayNumpy()
img = Image(res)
ellipse = self.ellipse
center = (img.width/2, img.height/2)
width = 10
diameter = int(max(ellipse.a, ellipse.b)*2)
mask = np.zeros_like(img.getGrayNumpy(), dtype=np.bool)
mask[center[0]-width/2:center[0]+width/2, center[1]-diameter/2:center[1]+diameter/2] = True
mask = Image(mask).threshold(0)
maxI = 0
maxThetaVal = 0
values = []
for i, theta in enumerate(xrange(-90, 90, 10)):
masked = (img & mask.rotate(theta, point=(img.width/2, img.height/2))).getGrayNumpy()
val = np.sum(masked)/float(img.width*img.height)
values.append(val)
if val > maxThetaVal:
maxThetaVal = val
maxI = i
result[0:len(values)-maxI] = values[maxI:]
result[len(values)-maxI:] = values[0:maxI]
self._rotBoxTest = result
return self._rotBoxTest
