Add the pfm visualized scripts

README.md

@@ -9,6 +9,7 @@ This is an unofficial Pytorch implementation of MVSNet
 ### Environment
 * python 3.6 (Anaconda)
 * pytorch 1.0.1
+* `pip install -r requirements.txt`
 
 ### Training
 
@@ -23,6 +24,14 @@ This is an unofficial Pytorch implementation of MVSNet
 * in ``test.sh``, set ``DTU_TESTING`` as your testing data path and ``CKPT_FILE`` as your checkpoint file. You can also download my [pretrained model](https://drive.google.com/file/d/1j2I_LNKb9JeCl6wdA7hh8z1WgVQZfLU9/view?usp=sharing).
 * Test MVSNet: ``./test.sh``
 
+### Depthmap Visualization
+
+<img src="doc/rgb.jpg" width="250">   | <img src="doc/depthmap.png" width="250"> |  <img src="doc/depthmap.jpg" width="250">
+:---------------------------------------:|:---------------------------------------:|:---------------------------------------:
+reference image                          |depth map (matplotlib)                   |  depth map (opencv) 
+
+Visualize the estimated depth map using `python visualize.py xxx.pfm`
+
 ### Fusion
 
 in ``eval.py``, I implemented a simple version of depth map fusion. Welcome contributions to improve the code.
@@ -36,3 +45,7 @@ in ``eval.py``, I implemented a simple version of depth map fusion. Welcome cont
 | PyTorch-MVSNet(D=192) | 0.4492 | 0.3796 | 0.4144   |
 
 Due to the memory limit, we only train the model with ``D=192``, the fusion code is also different from the original repo.
+
+### TODO
+1. add dataloader for [blendedmvs dataset](https://github.com/YoYo000/BlendedMVS)
+2. using [fusibile of mvsnet version](https://github.com/XYZ-qiyh/fusibile-mvsnet)

requirements.txt

@@ -0,0 +1,4 @@
+opencv-python
+plyfile
+matplotlib
+tensorboardX

tools/fusibile_to_dtu_eval.py

@@ -0,0 +1,36 @@
+# Convert output of fusibile to DTU evaluation format.
+# By: Jiayu Yang
+# Date: 2020-03-30
+
+import os
+from os import listdir
+
+fusibile_out_folder="../outputs-dtu/"
+dtu_eval_folder="../outputs-dtu/"
+
+if not os.path.isdir(dtu_eval_folder):
+    os.mkdir(dtu_eval_folder)
+
+scans = ["scan1", "scan4", "scan9", "scan10", "scan11",
+         "scan12", "scan13", "scan15", "scan23", "scan24",
+         "scan29", "scan32", "scan33", "scan34", "scan48",
+         "scan49", "scan62", "scan75", "scan77", "scan110",
+         "scan114", "scan118"]
+
+for scan in scans:
+    # Move ply to dtu eval folder and rename
+    scan_folder = os.path.join(fusibile_out_folder, scan, "points_mvsnet")
+    consis_folders = [f for f in listdir(scan_folder) if f.startswith('consistencyCheck-')]
+
+    consis_folders.sort()
+    consis_folder = consis_folders[-1]
+    source_ply = os.path.join(fusibile_out_folder, scan, "points_mvsnet", consis_folder, 'final3d_model.ply')
+    #print("source :{}".format(source_ply))
+    #source_ply = os.path.join(fusibile_out_folder,scan,'consistencyCheck/final3d_model.ply')
+    scan_idx = int(scan[4:])
+    target_ply = os.path.join(dtu_eval_folder,'mvsnet{:03d}_l3.ply'.format(scan_idx))
+
+    cmd = 'mv '+source_ply+' '+target_ply
+
+    print(cmd)
+    os.system(cmd)

tools/gray2color.py

@@ -0,0 +1,39 @@
+import numpy as np
+import cv2
+import argparse
+
+import sys
+sys.path.append("../datasets")
+from data_io import read_pfm#, write_depth_img
+
+
+def write_depth_img(filename, depth_image):
+    # Mask the array where equal to a given value
+    ma = np.ma.masked_equal(depth_image, 0.0, copy=False)
+    d_min = ma.min()
+    d_max = ma.max()
+    depth_n = 255.0 * (depth_image - d_min) / (d_max - d_min) # depth map normalize
+    depth_n = depth_n.astype(np.uint8)
+    out_depth_image = cv2.applyColorMap(depth_n, cv2.COLORMAP_JET) # applyColorMap
+    cv2.imwrite(filename, out_depth_image)    
+
+
+if __name__ == "__main__":
+    # parse argument
+    parser = argparse.ArgumentParser()
+    parser.add_argument("depth_path")
+    args = parser.parse_args()
+    depth_path = args.depth_path
+
+    # read_pfm 
+    depth_map, _ = read_pfm(depth_path)
+    print('depth shape: {}'.format(depth_map.shape))
+
+    ## photometric filter
+    #if False:
+    #    pfm_prob_path = depth_path.replace("depth_est", "confidence")
+    #    prob_map, _ = read_pfm(pfm_prob_path)
+    #    depth_map[prob_map < 0.9] = 0
+
+    # gray2color
+    write_depth_img(depth_path.replace(".pfm", ".jpg"), depth_map)

tools/visualize.py

@@ -0,0 +1,33 @@
+import numpy as np
+import cv2
+import argparse
+import matplotlib.pyplot as plt
+
+import sys
+sys.path.append("../datasets")
+from data_io import read_pfm
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('depth_path')
+    args = parser.parse_args()
+    depth_path = args.depth_path
+    if depth_path.endswith('npy'):
+        depth_image = np.load(depth_path)
+        depth_image = np.squeeze(depth_image)
+        print('value range: ', depth_image.min(), depth_image.max())
+        plt.imshow(depth_image, 'rainbow')
+        plt.show()
+    elif depth_path.endswith('pfm'):
+        depth_image, _ = read_pfm(depth_path)
+        print('depth shape: {}'.format(depth_image.shape))
+        ma = np.ma.masked_equal(depth_image, 0.0, copy=False)
+        print('value range: ', ma.min(), ma.max())
+        plt.imshow(depth_image, 'rainbow')
+        plt.show()
+    else:
+        depth_image = cv2.imread(depth_path)
+        ma = np.ma.masked_equal(depth_image, 0.0, copy=False)
+        print('value range: ', ma.min(), ma.max())
+        plt.imshow(depth_image)
+        plt.show()

tools/visualize_save.py

@@ -0,0 +1,36 @@
+import numpy as np
+import cv2
+import argparse
+import matplotlib.pyplot as plt
+
+import sys
+sys.path.append("../datasets")
+from data_io import read_pfm
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('depth_path')
+    args = parser.parse_args()
+    depth_path = args.depth_path
+    if depth_path.endswith('npy'):
+        depth_image = np.load(depth_path)
+        depth_image = np.squeeze(depth_image)
+        print('value range: ', depth_image.min(), depth_image.max())
+        plt.imshow(depth_image, 'rainbow')
+        plt.show()
+    elif depth_path.endswith('pfm'):
+        depth_image, _ = read_pfm(depth_path)
+        print('depth shape: {}'.format(depth_image.shape))
+        ma = np.ma.masked_equal(depth_image, 0.0, copy=False)
+        print('value range: ', ma.min(), ma.max())
+        plt.imshow(depth_image, 'rainbow')
+        # plt.show()
+        plt.axis('off')
+        out_filename = depth_path.replace(".pfm", ".png")
+        plt.savefig(out_filename, bbox_inches='tight', pad_inches=0, dpi=300)
+    else:
+        depth_image = cv2.imread(depth_path)
+        ma = np.ma.masked_equal(depth_image, 0.0, copy=False)
+        print('value range: ', ma.min(), ma.max())
+        plt.imshow(depth_image)
+        plt.show()