Merge branch 'master' of https://github.com/iitmcvg/CNN_SLAM

monodepth/monodepth_model.py

@@ -31,7 +31,7 @@
 class MonodepthModel(object):
     """monodepth model"""
 
-    def __init__(self, params, mode, left, right, reuse_variables=None, model_index=0):
+    def __init__(self, params, mode, left, right, reuse_variables=tf.AUTO_REUSE, model_index=0):
         self.params = params
         self.mode = mode
         self.left = left

monodepth_infer/monodepth_single.py

@@ -3,7 +3,7 @@
 import tensorflow.contrib.slim as slim
 import tensorflow as tf
 import numpy as np
-
+import cv2
 
 # Module imports
 import monodepth.average_gradients as average_gradients
@@ -46,7 +46,21 @@ def init_monodepth(checkpoint_path, sess = None):
     # SESSION
     if not sess:
         left  = tf.placeholder(tf.float32, [2, input_height, input_width, 3])
-        model = MonodepthModel(params, "test", left, None)
+        params = monodepth_model.monodepth_parameters(
+            encoder="resnet50",
+            height=input_height,
+            width=input_width,
+            batch_size=2,
+            num_threads=1,
+            num_epochs=1,
+            do_stereo=False,
+            wrap_mode="border",
+            use_deconv=False,
+            alpha_image_loss=0,
+            disp_gradient_loss_weight=0,
+            lr_loss_weight=0,
+            full_summary=False)
+        model = monodepth_model.MonodepthModel(params, "test", left, None)
         config = tf.ConfigProto(allow_soft_placement=True)
         sess = tf.Session(config=config)
 
@@ -81,7 +95,8 @@ def get_depth_map(sess, image_array, encoder="resnet50"):
     * Use frozen graphs instead
     * Switch to tf.data.Datasets
     '''
-    input_height, input_width = image_array.shape()[:2]
+    # org_input_height, org_input_width = image_array.shape[:2]
+
     params = monodepth_model.monodepth_parameters(
         encoder=encoder,
         height=input_height,
@@ -102,14 +117,14 @@ def get_depth_map(sess, image_array, encoder="resnet50"):
 
     input_image = image_array
     original_height, original_width, num_channels = input_image.shape
+    input_image = cv2.resize(input_image,(input_width, input_height),interpolation = cv2.INTER_CUBIC)
     input_image = input_image.astype(np.float32) / 255
     input_images = np.stack((input_image, np.fliplr(input_image)), 0)
 
     disp = sess.run(model.disp_left_est[0], feed_dict={left: input_images})
     disp_pp = post_process_disparity(disp.squeeze()).astype(np.float32)
 
-    disp_to_img = scipy.misc.imresize(
-        disp_pp.squeeze(), [
-            original_height, original_width])
+    disp_to_img = cv2.resize(
+        disp_pp.squeeze(), (original_height, original_width))
 
     return disp_to_img

run.py

@@ -19,7 +19,7 @@
 import pose_estimation.camera_pose_estimation as camera_pose_estimation
 import pose_estimation.find_uncertainty as find_uncertainty"""
 from keyframe_utils import Keyframe as Keyframe
-import monodepth_infer.monodepth_single as monodepth
+import monodepth_infer.monodepth_single as monodepth_single
 
 parser = argparse.ArgumentParser(description='Monodepth TensorFlow implementation.')
 parser.add_argument('--mono_checkpoint_path', default = "checkpoints/model_kitti_resnet/model_kitti_resnet.data" ,type=str,   help='path to a specific checkpoint to load')
@@ -84,8 +84,8 @@ def _exit_program():
 	raise NotImplementedError
 
 def main():
-	# INIT monodepth session
-	sess=monodepth.init_monodepth(args.mono_checkpoint_path)
+	# INIT monodepth_single session
+	sess=monodepth_single.init_monodepth(args.mono_checkpoint_path)
 
 	# INIT camera matrix
 	cam_matrix = get_camera_matrix()
@@ -103,7 +103,7 @@ def main():
 
 	# Predict depth
 	image = cv2.imread("pose_estimation/stereo.jpeg")
-	ini_depth = monodepth.get_cnn_depth(sess,image)
+	ini_depth = monodepth_single.get_depth_map(sess,image)
 	cv2.imshow('dawd',ini_depth)
 	cv2.waitKey(0)
 
@@ -131,7 +131,7 @@ def main():
 
 		if check_keyframe(T):			
 			# If it is a keyframe, add it to K after finding depth and uncertainty map                    
-			depth = monodepth.get_cnn_depth(sess,image)	
+			depth = monodepth_single.get_depth_map(sess,image)	
 			cur_index += 1
 			uncertainty = find_uncertainty.get_uncertainty(T,D,K[cur_index - 1])
 			# T = np.append(T,np.array([[0,0,0,1]]),0)
@@ -215,7 +215,7 @@ def test_without_cnn():
 	if check_keyframe(T):	
 		print ("Error: second frame cant be keyframe\n")	
 		# If it is a keyframe, add it to K after finding depth and uncertainty map                    
-		depth = monodepth.get_cnn_depth(sess,image)	
+		depth = monodepth_single.get_cnn_depth(sess,image)	
 		cur_index += 1
 		uncertainty = find_uncertainty.get_uncertainty(T,D,K[cur_index - 1])
 		T = np.append(T,np.array([[0,0,0,1]]),0)