HR_batch_dims = (1, height, width, 1) LR_batch_dims = (1, LR_height, LR_width, 1) #setting input placeholders HR_depth_batch_input = tf.placeholder(tf.float32, HR_batch_dims) HR_inten_batch_input = tf.placeholder(tf.float32, HR_batch_dims) LR_depth_batch_input = tf.placeholder(tf.float32, LR_batch_dims) coar_inter_dep_batch = tf.image.resize_images( LR_depth_batch_input, tf.constant(HR_patch_size, dtype=tf.int32), tf.image.ResizeMethod.BICUBIC) #gen_network construction inten_feature = bnb.inten_feature_extraction(HR_inten_batch_input) inten_1x_ten = bnb.inten_residual_block(inten_feature, 1) inten_ten, inten_2x_down = bnb.inten_residual_block(inten_1x_ten, 2) inten_2x_down_up = bnb.feature_up_unit(inten_2x_down, 1) dep_4x_ten = bnb.LR_dep_feature_extraction(LR_depth_batch_input) dep_4x_ten_up = bnb.feature_up_unit(dep_4x_ten, 2) dep_2x_ten = bnb.LR_dep_fusion(dep_4x_ten_up[0], inten_2x_down, fusion_stage=1) dep_2x_ten_up = bnb.feature_up_unit(dep_2x_ten, 1) dep_ten = bnb.LR_dep_fusion(tf.concat([dep_4x_ten_up[1], dep_2x_ten_up[0]], 3), tf.concat([inten_2x_down_up[0], inten_1x_ten], 3), fusion_stage=2) HR_gen_dep = bnb.LR_dep_reconstruction(dep_4x_ten_up[1], dep_ten, coar_inter_dep_batch) #define loss for gen saver_full = tf.train.Saver() #begin comp_gen testing with tf.Session() as sess:
HR_batch_dims = (1, height, width, 1) LR_batch_dims = (1, LR_height, LR_width, 1) #setting input placeholders HR_depth_batch_input = tf.placeholder(tf.float32, HR_batch_dims) HR_inten_batch_input = tf.placeholder(tf.float32, HR_batch_dims) LR_depth_batch_input = tf.placeholder(tf.float32, LR_batch_dims) coar_inter_dep_batch = tf.image.resize_images( LR_depth_batch_input, tf.constant(HR_patch_size, dtype=tf.int32), tf.image.ResizeMethod.BICUBIC) #gen_network construction inten_feature = bnb.inten_feature_extraction(HR_inten_batch_input) inten_1x_ten = bnb.inten_residual_block(inten_feature, 1) dep_2x_ten = bnb.LR_dep_feature_extraction(LR_depth_batch_input) dep_2x_ten_up = bnb.feature_up_unit(dep_2x_ten, 1) dep_ten = bnb.LR_dep_fusion(dep_2x_ten_up[0], inten_1x_ten, fusion_stage=1) HR_gen_dep = bnb.LR_dep_reconstruction(dep_2x_ten_up[0], dep_ten, coar_inter_dep_batch) #define loss for gen saver_full = tf.train.Saver() #begin comp_gen testing with tf.Session() as sess: model_path = "/media/kenny/Data/trained_models/multi_dense_guide_resnet/noise-free/l1loss/2x/full_model1/2x_ny_full_model.ckpt-99" saver_full.restore(sess, model_path) ten_fets = sess.run(HR_gen_dep, feed_dict={ HR_inten_batch_input: val_inten, HR_depth_batch_input: val_gth_dep,
#setting input size and training data addr HR_patch_size=[height,width] HR_batch_dims=(1,height,width,1) LR_batch_dims=(1,LR_height,LR_width,1) #setting input placeholders HR_depth_batch_input=tf.placeholder(tf.float32,HR_batch_dims) HR_inten_batch_input=tf.placeholder(tf.float32,HR_batch_dims) LR_depth_batch_input=tf.placeholder(tf.float32,LR_batch_dims) coar_inter_dep_batch=tf.image.resize_images(LR_depth_batch_input,tf.constant(HR_patch_size,dtype=tf.int32),tf.image.ResizeMethod.BICUBIC) #gen_network construction inten_feature=bnb.inten_feature_extraction(HR_inten_batch_input) inten_1x_ten=bnb.inten_residual_block(inten_feature,1) inten_ten,inten_2x_down=bnb.inten_residual_block(inten_1x_ten,2) inten_2x_down_up=bnb.feature_up_unit(inten_2x_down,1) inten_ten,inten_4x_down=bnb.inten_residual_block(inten_ten,4) inten_4x_down_up=bnb.feature_up_unit(inten_4x_down,2) inten_ten,inten_8x_down=bnb.inten_residual_block(inten_ten,8) inten_8x_down_up=bnb.feature_up_unit(inten_8x_down,3) dep_16x_ten=bnb.LR_dep_feature_extraction(LR_depth_batch_input) dep_16x_ten_up=bnb.feature_up_unit(dep_16x_ten,4) dep_8x_ten=bnb.LR_dep_fusion(dep_16x_ten_up[0],inten_8x_down,fusion_stage=1) dep_8x_ten_up=bnb.feature_up_unit(dep_8x_ten,3) dep_4x_ten=bnb.LR_dep_fusion(tf.concat([dep_16x_ten_up[1],dep_8x_ten_up[0]],3),tf.concat([inten_8x_down_up[0],inten_4x_down],3),fusion_stage=2) dep_4x_ten_up=bnb.feature_up_unit(dep_4x_ten,2) dep_2x_ten=bnb.LR_dep_fusion(tf.concat([dep_16x_ten_up[2],dep_8x_ten_up[1],dep_4x_ten_up[0]],3),tf.concat([inten_8x_down_up[1],inten_4x_down_up[0],inten_2x_down],3),fusion_stage=3) dep_2x_ten_up=bnb.feature_up_unit(dep_2x_ten,1) dep_ten=bnb.LR_dep_fusion(tf.concat([dep_16x_ten_up[3],dep_8x_ten_up[2],dep_4x_ten_up[1],dep_2x_ten_up[0]],3),tf.concat([inten_8x_down_up[2],inten_4x_down_up[1],inten_2x_down_up[0],inten_1x_ten],3),fusion_stage=4) HR_gen_dep=bnb.LR_dep_reconstruction(dep_16x_ten_up[3],dep_ten,coar_inter_dep_batch)