def __init__(self,
weight_fn,
camK,
res_x,
res_y,
obj_param,
th_ransac=3.0,
th_outlier=[0.1, 0.2, 0.3],
th_inlier=0.1,
box_size=1.5,
dist_coeff=None,
backbone="paper",
**kwargs):
self.camK = camK
self.res_x = res_x
self.res_y = res_y
self.th_ransac = th_ransac
self.th_o = th_outlier
self.th_i = th_inlier
self.obj_scale = obj_param[:3] #x,y,z
self.obj_ct = obj_param[3:] #x,y,z
self.box_size = box_size
self.dist_coeff = dist_coeff
if (backbone == 'paper'):
self.generator_train = ae.aemodel_unet_prob(p=1.0) #output:3gae
self.generator_train.load_weights(weight_fn)
elif (backbone == 'resnet50'):
self.generator_train = load_model(weight_fn)
if (fn_temp.startswith("pix2pose" + ".")
and fn_temp.endswith("hdf5")):
temp_split = fn_temp.split(".")
epoch_split = temp_split[1].split("-")
epoch_split2 = epoch_split[0].split("_")
epoch_temp = int(epoch_split2[0])
if (epoch_temp > recent_epoch):
recent_epoch = epoch_temp
weight_fn = fn_temp
if os.path.exists(os.path.join(root,
"inference.hdf5")) and pass_exists:
print("A converted file exists in ",
os.path.join(root, "inference.hdf5"))
continue
if (weight_fn != ""):
generator_train = ae.aemodel_unet_prob(p=1.0)
discriminator = ae.DCGAN_discriminator()
imsize = 128
dcgan_input = Input(shape=(imsize, imsize, 3))
dcgan_target = Input(shape=(imsize, imsize, 3))
prob_gt = Input(shape=(imsize, imsize, 1))
gen_img, prob = generator_train(dcgan_input)
recont_l = ae.transformer_loss(
[np.eye(3)])([gen_img, dcgan_target, prob_gt, prob_gt])
disc_out = discriminator(gen_img)
dcgan = Model(inputs=[dcgan_input, dcgan_target, prob_gt],
outputs=[recont_l, disc_out, prob])
print("load recent weights from ", os.path.join(root, weight_fn))
dcgan.load_weights(os.path.join(root, weight_fn))
print("save recent weights to ",