new depth model

gimpml/plugins/monodepth/monodepth.py

@@ -121,8 +121,6 @@ def monodepth(procedure, image, drawable, force_cpu, progress_bar, config_path_o
         copy.set_mode(Gimp.LayerMode.NORMAL_LEGACY)  # DIFFERENCE_LEGACY
         image.insert_layer(copy, None, -1)
 
-
-
         # Remove temporary layers that were saved
         my_dir = os.path.join(weight_path, '..')
         for f_name in os.listdir(my_dir):
@@ -191,17 +189,17 @@ def run(procedure, run_mode, image, n_drawables, layer, args, data):
         vbox.pack_start(label, False, False, 1)
         label.show()
 
-        # Show ideal image size text
-        label = Gtk.Label(label="384 X 384 px | ")
-        grid.attach(label, 1, 0, 1, 1)
-        # vbox.pack_start(label, False, False, 1)
-        label.show()
+        # # Show ideal image size text
+        # label = Gtk.Label(label="384 X 384 px | ")
+        # grid.attach(label, 1, 0, 1, 1)
+        # # vbox.pack_start(label, False, False, 1)
+        # label.show()
 
         # Force CPU parameter
         spin = GimpUi.prop_check_button_new(config, "force_cpu", _("Force _CPU"))
         spin.set_tooltip_text(_("If checked, CPU is used for model inference."
                                 " Otherwise, GPU will be used if available."))
-        grid.attach(spin, 2, 0, 1, 1)
+        grid.attach(spin, 1, 0, 1, 1)
         spin.show()
 
         progress_bar = Gtk.ProgressBar()

gimpml/tools/model_info.csv

@@ -1,6 +1,8 @@
 model,file_id,fileSize,mFName,md5sum
 deepmatting,11dxJKH8p7xkcGtMtvzMUw-ua6pZ0vrfw,108,stage1_sad_57.1.pth,0cffb16b30fad67fa3eb5cfeeae0c7d5
 MiDaS,11eap5jc-4SCX_sMMxYE6Bi5q_BKw894a,143,model.pt,74a1fcdde0b743212851d4b81be5f6eb
+MiDaS,1hmv768Mrfv56SitcoMUlYp6zhUrj0HQv,493,dpt_hybrid-midas-501f0c75.pt,39beff6ce86ce03f6b94a29c6d11ee12
+semseg,19rLDu8COUFSrOMnk124SBiEN9S9_QHBH,496,dpt_hybrid-ade20k-53898607.pt,d59f62cfba50297ffdef053bc4818a21
 colorize,12tKfNIDewgJPbW3FiITV_AMbOtZWP0Eg,130,caffemodel.pth,efb7de5b0f3827d8a7e392ac287bca81
 superresolution,11GwnqKsYo2jujACD_GMB9uMTQfsuk2RY,6,model_srresnet.pth,8f7ed4feb00402863e771a46013a93a8
 faceparse,115nnWD0FoDkplTJYBY7lTQu1VNXFbCA_,51,79999_iter.pth,ff26a222ce48a618a1fa820b46223cae

gimpml/tools/monodepth.py

@@ -1,13 +1,15 @@
 import pickle
 import os
 import sys
+import traceback
 
 plugin_loc = os.path.dirname(os.path.realpath(__file__)) + '/'
-sys.path.extend([plugin_loc + 'MiDaS'])
+sys.path.extend([plugin_loc + 'DPT'])
 
-from mono_run import run_depth
-from monodepth_net import MonoDepthNet
-import MiDaS_utils as MiDaS_utils
+from monodepth_run import run
+# from mono_run import run_depth
+# from monodepth_net import MonoDepthNet
+# import MiDaS_utils as MiDaS_utils
 import numpy as np
 import cv2
 import torch
@@ -21,13 +23,27 @@ def get_weight_path():
     return weight_path
 
 
+# def get_mono_depth(input_image, cpu_flag=False, weight_path=None):
+#     if weight_path is None:
+#         weight_path = get_weight_path()
+#     image = input_image / 255.0
+#     with torch.no_grad():
+#         out = run_depth(image, os.path.join(weight_path, 'MiDaS', 'model.pt'), MonoDepthNet, MiDaS_utils, target_w=640,
+#                         f=cpu_flag)
+#     out = np.repeat(out[:, :, np.newaxis], 3, axis=2)
+#     d1, d2 = input_image.shape[:2]
+#     out = cv2.resize(out, (d2, d1))
+#     return out
+
+
 def get_mono_depth(input_image, cpu_flag=False, weight_path=None):
     if weight_path is None:
         weight_path = get_weight_path()
-    image = input_image / 255.0
+
     with torch.no_grad():
-        out = run_depth(image, os.path.join(weight_path, 'MiDaS', 'model.pt'), MonoDepthNet, MiDaS_utils, target_w=640,
-                        f=cpu_flag)
+        out = run(input_image, os.path.join(weight_path, 'MiDaS', 'dpt_hybrid-midas-501f0c75.pt'), cpu_flag=cpu_flag, bits=2)
+        # out = run_depth(image, os.path.join(weight_path, 'MiDaS', 'model.pt'), MonoDepthNet, MiDaS_utils, target_w=640,
+        #                 f=cpu_flag)
     out = np.repeat(out[:, :, np.newaxis], 3, axis=2)
     d1, d2 = input_image.shape[:2]
     out = cv2.resize(out, (d2, d1))
@@ -42,7 +58,11 @@ if __name__ == "__main__":
     force_cpu = data_output["force_cpu"]
     try:
         output = get_mono_depth(image, cpu_flag=force_cpu, weight_path=weight_path)
-        cv2.imwrite(os.path.join(weight_path, '..', 'cache.png'), output[:, :, ::-1])
+        # if bits == 1:
+        #     cv2.imwrite(os.path.join(weight_path, '..', 'cache.png'), output.astype("uint8"), [cv2.IMWRITE_PNG_COMPRESSION, 0])
+        # elif bits == 2:
+        cv2.imwrite(os.path.join(weight_path, '..', 'cache.png'), output.astype("uint16"), [cv2.IMWRITE_PNG_COMPRESSION, 0])
+        # cv2.imwrite(os.path.join(weight_path, '..', 'cache.png'), output[:, :, ::-1])
         with open(os.path.join(weight_path, '..', 'gimp_ml_run.pkl'), 'wb') as file:
             pickle.dump({"inference_status": "success", "force_cpu": force_cpu}, file)
 
@@ -54,5 +74,8 @@ if __name__ == "__main__":
     except Exception as error:
         with open(os.path.join(weight_path, '..', 'gimp_ml_run.pkl'), 'wb') as file:
             pickle.dump({"inference_status": "failed"}, file)
-        with open(os.path.join(weight_path, '..', 'error_log.txt'), 'w') as file:
-            file.write(str(error))
+        # with open(os.path.join(weight_path, '..', 'error_log.txt'), 'w') as file:
+        #     file.write(str(error))
+        with open(os.path.join(weight_path, '..', 'error_log.txt'), 'w') as f:
+            f.write(str(error))
+            f.write(traceback.format_exc())

setup.py

@@ -70,7 +70,7 @@ setup(
     # https://packaging.python.org/en/latest/requirements.html
     install_requires=['numpy', 'future; python_version <= "2.7"',
                       'scipy', 'gdown', 'typing', 'enum; python_version <= "2.7"', 'requests', 'opencv-python<=4.3',
-                      'pretrainedmodels', "scikit-image"]#, "torch==1.8", "torchvision"],
+                      'pretrainedmodels', "scikit-image", "timm==0.4.5"]#, "torch==1.8", "torchvision"],
 
 
     # Optional