[MOD] evaluate_depth_densification_by... files

tensorflow/evaluate_depth_densification_by_close_operation.py

@@ -0,0 +1,150 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+# ===========
+#  Libraries
+# ===========
+import os
+import sys
+import time
+import warnings
+
+import imageio
+import matplotlib.pyplot as plt
+import numpy as np
+from skimage import exposure, img_as_uint
+
+# Custom Libraries
+from modules import metrics
+from modules.args import args
+from modules.utils import settings
+import cv2
+
+# =========================
+#  [Test] Framework Config
+# =========================
+SAVE_TEST_DISPARITIES = True  # Default: True
+showImages = True
+
+# ==================
+#  Global Variables
+# ==================
+os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu  # Limits TensorFlow to see only the specified GPU.
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
+warnings.filterwarnings("ignore")  # Suppress Warnings
+
+
+def read_text_file(filename, dataset_path='/media/nicolas/nicolas_seagate/datasets/kitti/'):
+
+    print("\n[Dataloader] Loading '%s'..." % filename)
+    try:
+        data = np.genfromtxt(filename, dtype='str', delimiter='\t')
+        # print(data.shape)
+
+        # Parsing Data
+        depth_continuous_filenames = list(data[:, 0])
+        depth_semidense_filenames = list(data[:, 1])
+
+        timer = -time.time()
+        depth_continuous_filenames = [dataset_path + filename for filename in depth_continuous_filenames]
+        depth_semidense_filenames = [dataset_path + filename for filename in depth_semidense_filenames]
+        timer += time.time()
+        print('time:', timer, 's\n')
+
+    except OSError:
+        raise OSError("Could not find the '%s' file." % filename)
+
+    return depth_continuous_filenames, depth_semidense_filenames
+
+def read_depth_image(filename, div=1.0):
+    return imageio.imread(filename).astype('float32') / div
+
+
+def imsave_as_uint16_png(filename, image_float32):
+    # Converts the Predictions Images from float32 to uint16 and Saves as PNG Images
+
+    image_uint16 = img_as_uint(exposure.rescale_intensity(image_float32, out_range='float'))
+    imageio.imsave(filename, image_uint16)
+
+
+def evaluate_densification():
+    # Loads split file containing Input and Output filenames
+    # input_filenames, output_filenames = read_text_file('data/new_splits/eigen_split_based_on_kitti_depth/eigen_test_kitti_depth_aligned_with_kitti_continuous_files.txt')
+    input_filenames = ['/home/nicolas/Downloads/depth_interpolation/close/0000000005_close_k_2.png']
+    output_filenames = ['/home/nicolas/Downloads/depth_interpolation/close/0000000005.png']
+
+    assert len(input_filenames) == len(output_filenames)
+    print(len(input_filenames), len(output_filenames))
+
+    # Read Images
+    input_depths, output_depths = [], []
+    num_test_images = len(output_filenames)
+
+    fig, (ax1, ax2, ax3, ax4, ax5) = plt.subplots(5, 1)
+    for i, (input_filename, output_filename) in enumerate(list(zip(input_filenames, output_filenames))):
+        close_depth = read_depth_image(input_filename, 256.0)
+        kitti_depth_depth = read_depth_image(output_filename, 256.0)
+
+        # Fix Data shift caused by close operation
+        # TODO: Esta correção só precisa ser feita se o kernel utilizado no close é par.
+        rows, cols = close_depth.shape
+        tx, ty = -1, -1 # Offsets
+        M = np.float32([[1, 0, tx], [0, 1, ty]])
+        close_depth = cv2.warpAffine(close_depth, M, (cols, rows))
+
+        artefacts = close_depth - kitti_depth_depth
+        real_proof = kitti_depth_depth+artefacts
+        real_proof2 = (close_depth - real_proof)*1000
+        print(real_proof2)
+
+        print(close_depth[close_depth > 0.0].size,
+              kitti_depth_depth[kitti_depth_depth > 0.0].size,
+              artefacts[artefacts > 0.0].size)  # Number of Valid Pixels
+
+        print(close_depth.shape)
+        print(kitti_depth_depth.shape)
+        print(np.min(close_depth), np.max(close_depth))
+
+        if showImages:
+            ax1.imshow(close_depth)
+            ax1.set_title('close(k=2)')
+            ax2.imshow(kitti_depth_depth)
+            ax2.set_title('KITTI Depth')
+            ax3.imshow(artefacts)
+            ax3.set_title('Artefacts')
+            ax4.imshow(real_proof)
+            ax4.set_title('KITTI Depth + Artefacts')
+            ax5.imshow(real_proof2)
+            ax5.set_title('(KITTI Depth + Artefacts)-close(k=2)')
+
+            # plt.draw()
+            # plt.pause(0.001)
+            plt.show()
+
+        # Saves the Test Predictions as uint16 PNG Images
+        if SAVE_TEST_DISPARITIES or args.eval_tool == 'monodepth':
+            imsave_as_uint16_png(settings.output_tmp_pred_dir + 'pred' + str(i) + '.png', close_depth)
+            imsave_as_uint16_png(settings.output_tmp_gt_dir + 'gt' + str(i) + '.png', kitti_depth_depth)
+
+        input_depths.append(close_depth)
+        output_depths.append(kitti_depth_depth)
+
+        print('{}/{}'.format(i+1, num_test_images))
+
+
+    # Invokes Evaluation Tools
+    if args.eval_tool == 'monodepth':
+        metrics.evaluation_tool_monodepth(input_depths, output_depths)
+    elif args.eval_tool == 'kitti_depth':
+        metrics.evaluation_tool_kitti_depth(num_test_images)
+    else:
+        raise SystemError("Invalid 'eval_tool' selected. Choose one of the options: 'monodepth' or 'kitti_depth'.")
+
+
+# ======
+#  Main
+# ======
+if __name__ == '__main__':
+    evaluate_densification()
+
+    print("\nDone.")
+    sys.exit()