def imagine2(dim, code, save=False, visualize=True):
# set the shape
shape = (1, dim, dim, zed)
code1 = code[0:zed]
code2 = code[zed:zed * 2]
# overall tone
means = code1 - 0.5 # -0.5 or 0.5
variances = code2 * .3 + .2 # .2 or .5
# final latent
z = np.random.normal(loc=means, scale=variances, size=shape)
gened = gm.infer(z)
gened *= 0.5 # normalize to 0..1
gened += 0.5
im = gened[0]
if visualize:
# visualize
vis.show_autoscaled(im, name='imagined', limit=600.)
fmap = np.transpose(z, (3, 1, 2, 0)) + .5
vis.show_batch_autoscaled(fmap, name='z_map', limit=600.)
# outputs the image
return im
def process(self,lastframe,thisframe):
# graify, downscale
fg = thisframe.copy()
# fg = cv2.cvtColor(fg, cv2.COLOR_BGR2GRAY)
pyramid_downscaler = 2
pyramid = []
for i in range(pyramid_downscaler):
fg = pyr(fg,-1)
pyramid.append(fg)
# size: -1 -2 -3
probmap = None
for idx,img in enumerate(pyramid):
cv2.imshow('pyr'+str(idx), img)
# inferrence
img.shape = (1,)+img.shape
img = img.astype('float32')
j = clf.infer(img/255.-0.5)[0]
# j = j * (j>0.5)
# vis.show_autoscaled(j[0],name='infer_pyr_'+str(idx),limit=300.)
j = pyr(j, idx) # idx: [0 1 2]
if j is None:
print('j is none',img.dtype)
if probmap is None:
probmap = j
else:
if len(j.shape)==2:
j.shape+=(1,)
offy = int((probmap.shape[0] - j.shape[0])//2)
offx = int((probmap.shape[1] - j.shape[1])//2)
probmap[offy:offy+j.shape[0],offx:offx+j.shape[1],:] += j
# cv2.imshow('infer_pyr_1',j[0])
# probmap = cv2.blur(probmap,(3,3))
(minVal, maxVal, minLoc, maxLoc) = cv2.minMaxLoc(probmap)
# image = orig.copy()
if maxVal>0.9:
shift = 50 # a small offset since convnet cut the border off
scaler = (thisframe.shape[1] - shift*2) / probmap.shape[1]
# black-white ring
cv2.circle(thisframe,
(int(maxLoc[0]*scaler+shift),int(maxLoc[1]*scaler+shift)),
30, (0, 0, 0), 4)
cv2.circle(thisframe,
(int(maxLoc[0]*scaler+shift),int(maxLoc[1]*scaler+shift)),
30, (255, 255, 255), 2)
vis.show_autoscaled(probmap,name='scaled_back',limit=300.)
return thisframe
def qrtest():
uvout, qrout = random_qr_pipeline(128, dirty=True)
print('uvout', uvout.shape, 'qrout', qrout.shape)
# qrimg, uvimg = [pipeline(i) for i in [qrimg,uvimg]]
vis.show_autoscaled(qrout, name='qr')
vis.show_autoscaled(uvout, name='uv')
def show(save=False):
from cv2tools import vis, filt
i = np.random.normal(loc=0., scale=1., size=(1, 8, 8, zed))
gened = gm.infer(i)
gened *= 0.5
gened += 0.5
vis.show_autoscaled(gened[0], name='generated', limit=800)
if save != False:
cv2.imwrite(save, im * 255)
def compose_batch(self, output_size, batch_size, show=True):
bs = batch_size
bgrw, bgs = random_walk(bs, ipos=.5, ispd=5.,
acc=1.) #rw position and speed
fgrw, fgs = random_walk(bs, ipos=.5, ispd=5., acc=2.)
bgrw -= np.mean(bgrw) # normalization
fgrw -= np.mean(fgrw) # +np.random.uniform(40)-20 # normalization
batch_img = []
batch_gt = []
fg_gamma = rn(1) + .4
fg_beta = .25 - rn(.5)
bg_gamma = rn(1) + .4
bg_beta = .25 - rn(.5)
has_fg = rn(1) > 0.5
for i in range(len(bgrw)):
if True: # test flag
img, gt = self.compose_one(output_size,
fgrw[i],
bgrw[i],
fg_blur=fgs[i] / 2,
bg_blur=bgs[i] / 2,
fg_gamma=fg_gamma,
bg_gamma=bg_gamma,
fg_beta=fg_beta,
bg_beta=bg_beta,
has_fg=has_fg)
else:
img, gt = self.compose_one(output_size,
fgrw[i] * 0,
bgrw[i] * 0,
fg_blur=fgs[i] / 2 * 0,
bg_blur=bgs[i] / 2 * 0)
batch_img.append(img)
batch_gt.append(gt)
if show == True:
for k in range(len(batch_img)):
vis.show_autoscaled(batch_img[k], name='img', limit=130.)
vis.show_autoscaled(batch_gt[k], name='gt', limit=130.)
# both generated and ground truth
return np.stack(batch_img, axis=0), np.stack(batch_gt, axis=0)
def jeffdemo():
jeff = load_image('../jeff.jpg')
h, w = jeff.shape[0:2]
jeff = vis.resize_perfect(jeff, 192, 192)
print(jeff.shape)
vis.show_autoscaled(jeff)
jefff = np.divide(jeff, 255., dtype=np.float32)
jeffff = forenet_infer(jefff)
print(jeffff.shape)
jeffff = np.transpose(jeffff, [2, 0, 1])
jeffff.shape += (1, )
vis.show_batch_autoscaled(jeffff * 0.5 + 0.5)
cv2.waitKey(0)
def r(ep):
for i in range(ep):
env.step()
if (i + 1) % 5 == 0:
epop = env.population
best = epop[-1]
print('iter {}/{} '.format(i+1, ep)+\
'max:{:.6f} min:{:.6f} avg:{:.6f} avg_len:{:.1f}'.format(
epop[-1].fitness,
epop[0].fitness,
env.get_mean_fitness(),
env.get_mean_length(),
)+' popsize:{}'.format(len(epop))\
+'params:[{}]'.format(best.points[0])
)
c = env.get_blank_canvas()
env.population[-1].draw_on(c)
vis.show_autoscaled(c, name='best')
c = env.get_blank_canvas()
env.population[0].draw_on(c)
vis.show_autoscaled(c, name='worst')
vis.show_autoscaled(env.target, name='targ')
cv2.waitKey(1)
def render(self):
if self.target_drawn == False:
vis.show_autoscaled(self.target,limit=300,name='target')
self.target_drawn = True
vis.show_autoscaled(self.canvas,limit=300,name='canvas')
def show():
vis.show_autoscaled(jeff,name='jeff')
canvas = getcanvas()
vis.show_autoscaled(canvas)
cv2.waitKey(1)
