z_show = sample_Z(3, Z_dim)
y_show = show_Y(3)
magnify_interval = False
for it in range(1000000):
if it % 100 == 0:
samples = sess.run(G_sample, feed_dict={Y: y_show, Z: z_show})
save_img(dir=dir_name,
iter=it,
imgs=y_show,
boxs=samples,
magnify_interval=magnify_interval)
Y_mb, X_mb = get_batch(mb_size,
crop_img=False,
magnify_interval=magnify_interval)
# X_mb = np.reshape(X_mb, [-1, 28, 28, 1])
_, D_loss_curr = sess.run([D_optimizer, D_loss],
feed_dict={
X: X_mb,
Y: Y_mb,
Z: sample_Z(mb_size, Z_dim)
})
_, G_loss_curr = sess.run([G_optimizer, G_loss],
feed_dict={
Y: Y_mb,
Z: sample_Z(mb_size, Z_dim)
})
if it % 100 == 0:
def show_Y(m):
Y_mb, _ = get_batch(m, crop_img=False, magnify_interval=False)
return Y_mb
i = 0
mb_size = 128
Z_dim = 100
z_show = sample_Z(16, Z_dim)
for it in range(1000000):
if it % 100 == 0:
samples = sess.run(G_sample, feed_dict={Z: z_show})
fig = plot(samples)
plt.savefig(dir_name + '/{}.png'.format(str(i).zfill(3)),
bbox_inches='tight')
i += 1
plt.close(fig)
X_mb = get_batch(mb_size)
# X_mb = np.reshape(X_mb, [-1, 28, 28, 1])
_, D_loss_curr = sess.run([D_optimizer, D_loss],
feed_dict={
X: X_mb,
Z: sample_Z(mb_size, Z_dim)
})
_, G_loss_curr = sess.run([G_optimizer, G_loss],
feed_dict={Z: sample_Z(mb_size, Z_dim)})
if it % 100 == 0:
print('Iter: {}'.format(it))
print('D loss: {:.4}'.format(D_loss_curr))
print('G_loss: {:.4}'.format(G_loss_curr))
print()
Z_dim = 100
z_show = sample_Z(16, Z_dim)
for it in range(1000000):
if it % 100 == 0:
samples = sess.run(G_sample, feed_dict={Z: z_show})
samples = (samples +
1.) / 2. # inverse transform from [-1,1] to [0,1]
# fig = plot(samples)
# plt.savefig(dir_name + '/{}.png'.format(str(i).zfill(3)), bbox_inches='tight')
i += 1
# plt.close(fig)
from scipy import misc
misc.imsave(dir_name + '/{}.png'.format(str(i)), samples[0])
X_mb, _ = get_batch(mb_size, crop_img=False, magnify_interval=True)
# X_mb = np.reshape(X_mb, [-1, 28, 28, 1])
_, D_loss_curr = sess.run([D_optimizer, D_loss],
feed_dict={
X: X_mb,
Z: sample_Z(mb_size, Z_dim)
})
_, G_loss_curr = sess.run([G_optimizer, G_loss],
feed_dict={Z: sample_Z(mb_size, Z_dim)})
if it % 100 == 0:
print('Iter: {}'.format(it))
print('D loss: {:.4}'.format(D_loss_curr))
print('G_loss: {:.4}'.format(G_loss_curr))
print()
cv2.rectangle(img, (xmin, ymin), (xmax, ymax), (0, 255, 0), 2)
cv2.imwrite('{}/{}_{}.png'.format(dir, j, iter), img)
j += 1
else:
for i, img in enumerate(imgs):
img = img * 255
img = RGB2BGR(img)
cv2.imwrite('{}/{}_{}.png'.format(dir, i, iter), img)
if __name__ == '__main__':
from datas import get_batch
batch_size = 4
imgs_batch, box_batch = get_batch(batch_size,
crop_img=False,
magnify_interval=False)
print(len(box_batch))
# save_img('test', 2, imgs_batch)
# import cv2
# from scipy import misc
# misc.imsave('001_misc.png', imgs_batch[0])
# img = imgs_batch[0] * 255
# img = img[..., ::-1]
# img = img.copy()
# cv2.rectangle(img, (212, 300), (290, 300), (0, 255, 0), 2)
# cv2.imwrite('001_cv.png', img)
save_img('test', 1, imgs_batch, box_batch)
save_img('test', 2, imgs_batch)
mb_size = 32
Z_dim = 100
z_show = sample_Z(16, Z_dim)
for it in range(1000000):
if it % 100 == 0:
samples = sess.run(G_sample, feed_dict={Z: z_show})
samples = (samples +
1.) / 2. # inverse transform from [-1,1] to [0,1]
fig = plot(samples)
plt.savefig(dir_name + '/{}.png'.format(str(i).zfill(3)),
bbox_inches='tight')
i += 1
plt.close(fig)
X_mb = get_batch(mb_size, magnify_interval=True)
# X_mb = np.reshape(X_mb, [-1, 28, 28, 1])
_, D_loss_curr = sess.run([D_optimizer, D_loss],
feed_dict={
X: X_mb,
Z: sample_Z(mb_size, Z_dim)
})
_, G_loss_curr = sess.run([G_optimizer, G_loss],
feed_dict={Z: sample_Z(mb_size, Z_dim)})
if it % 100 == 0:
print('Iter: {}'.format(it))
print('D loss: {:.4}'.format(D_loss_curr))
print('G_loss: {:.4}'.format(G_loss_curr))
print()
mb_size = 128
Z_dim = 100
z_show = sample_Z(16, Z_dim)
for it in range(1000000):
if it % 100 == 0:
samples = sess.run(G_sample, feed_dict={Z: z_show})
samples = (samples +
1.) / 2. # inverse transform from [-1,1] to [0,1]
fig = plot(samples)
plt.savefig(dir_name + '/{}.png'.format(str(i).zfill(3)),
bbox_inches='tight')
i += 1
plt.close(fig)
X_mb = get_batch(mb_size, resize=True)
# X_mb = np.reshape(X_mb, [-1, 28, 28, 1])
_, D_loss_curr = sess.run([D_optimizer, D_loss],
feed_dict={
X: X_mb,
Z: sample_Z(mb_size, Z_dim)
})
_, G_loss_curr = sess.run([G_optimizer, G_loss],
feed_dict={Z: sample_Z(mb_size, Z_dim)})
if it % 100 == 0:
print('Iter: {}'.format(it))
print('D loss: {:.4}'.format(D_loss_curr))
print('G_loss: {:.4}'.format(G_loss_curr))
print()