fix ConverterMasked.py,

changing requirements
changing device.py ENV vars

converters/ConverterMasked.py

@@ -136,9 +136,15 @@ class ConverterMasked(Converter):
             debugs += [img_face_mask_aaa.copy()]
    
         if 'seamless' in self.mode:
-            img_face_seamless_mask_aaa = img_face_mask_aaa.copy()   #mask used for cv2.seamlessClone
+            #mask used for cv2.seamlessClone
-            img_face_seamless_mask_aaa[img_face_seamless_mask_aaa > 0.9] = 1.0
+            img_face_seamless_mask_aaa = None
-            img_face_seamless_mask_aaa[img_face_seamless_mask_aaa <= 0.9] = 0.0
+            for i in range(9, 0, -1):
+                a = img_face_mask_aaa > i / 10.0
+                if len(np.argwhere(a)) == 0:
+                    continue
+                img_face_seamless_mask_aaa = img_face_mask_aaa.copy()
+                img_face_seamless_mask_aaa[a] = 1.0
+                img_face_seamless_mask_aaa[img_face_seamless_mask_aaa <= i / 10.0] = 0.0
         
         out_img = img_bgr.copy()
         
@@ -155,8 +161,7 @@ class ConverterMasked(Converter):
             if self.raw_mode == 'predicted-only':
                 out_img = cv2.warpAffine( prd_face_bgr, face_output_mat, img_size, np.zeros(out_img.shape, dtype=np.float32), cv2.WARP_INVERSE_MAP | cv2.INTER_LANCZOS4, cv2.BORDER_TRANSPARENT )
                 
-        else:
+        elif ('seamless' not in self.mode) or (img_face_seamless_mask_aaa is not None):            
-
             #averaging [lenx, leny, maskx, masky] by grayscale gradients of upscaled mask
             ar = []
             for i in range(1, 10):                

nnlib/device.py

@@ -3,10 +3,9 @@ import json
 import numpy as np
 from .pynvml import *
 
- 
+#you can set DFL_TF_MIN_REQ_CAP manually for your build
-#you can force_tf_min_req_cap 35, if your DFL is built for tf==1.5.0
 #the reason why we cannot check tensorflow.version is it requires import tensorflow
-tf_min_req_cap = int(os.environ.get("force_tf_min_req_cap", 37)) 
+tf_min_req_cap = int(os.environ.get("DFL_TF_MIN_REQ_CAP", 35)) 
     
 class device:
     backend = None
@@ -260,14 +259,15 @@ class device:
         return result[0] * 10 + result[1]
 
         
-force_plaidML = os.environ.get("force_plaidML", "0") == "1"
+force_plaidML = os.environ.get("DFL_FORCE_PLAIDML", "0") == "1" #for OpenCL build , forcing using plaidML even if NVIDIA found
+force_tf_cpu = os.environ.get("DFL_FORCE_TF_CPU", "0") == "1"   #for OpenCL build , forcing using tf-cpu if plaidML failed
 has_nvml = False
 has_nvml_cap = False
 
-#use force_has_nvidia_device=1 if 
+#use DFL_FORCE_HAS_NVIDIA_DEVICE=1 if 
 #- your NVIDIA cannot be seen by OpenCL
 #- CUDA build of DFL
-has_nvidia_device = os.environ.get("force_has_nvidia_device", "0") == "1" 
+has_nvidia_device = os.environ.get("DFL_FORCE_HAS_NVIDIA_DEVICE", "0") == "1" 
 
 plaidML_devices = []
 
@@ -294,7 +294,7 @@ plaidML_devices_count = len(plaidML_devices)
 
 #choosing backend
 
-if device.backend is None:
+if device.backend is None and not force_tf_cpu:
     #first trying to load NVSMI and detect CUDA devices for tensorflow backend,
     #even force_plaidML is choosed, because if plaidML will fail, we can choose tensorflow
     try:
@@ -320,13 +320,15 @@ if device.backend is None:
 if not has_nvidia_device and (device.backend is None or force_plaidML):
     #tensorflow backend was failed without has_nvidia_device , or forcing plaidML, trying to use plaidML backend
     if plaidML_devices_count == 0:
-        print ("plaidML: No capable OpenCL devices found. Falling back to tensorflow backend.")
+        #print ("plaidML: No capable OpenCL devices found. Falling back to tensorflow backend.")
         device.backend = None
     else:
         device.backend = "plaidML"
 
 if device.backend is None:
-    if not has_nvml:        
+    if force_tf_cpu:
+        device.backend = "tensorflow-cpu"
+    elif not has_nvml:        
         if has_nvidia_device:
             #some notebook systems have NVIDIA card without NVSMI in official drivers
             #in that case considering we have system with one capable GPU and let tensorflow to choose best GPU

requirements-SSE-cuda9.0-cudnn7.1.4.txt

@@ -1,13 +0,0 @@
-numpy==1.16.1
-pathlib==1.0.1
-scandir==1.6
-h5py==2.7.1
-Keras==2.2.4
-opencv-python==4.0.0.21
-tensorflow-gpu==1.5.0
-plaidml-keras==0.5.0
-scikit-image
-dlib==19.10.0
-tqdm
-ffmpeg-python==0.1.17
-git+https://www.github.com/keras-team/keras-contrib.git

requirements-cpu.txt

@@ -1,10 +1,8 @@
 numpy==1.16.1
-pathlib==1.0.1
+h5py==2.9.0
-scandir==1.6
-h5py==2.7.1
 Keras==2.2.4
 opencv-python==4.0.0.21
-tensorflow==1.13.1
+tensorflow==1.12.0
 scikit-image
 dlib==19.16.0
 tqdm

requirements-cuda.txt

@@ -1,10 +1,8 @@
 numpy==1.16.1
-pathlib==1.0.1
+h5py==2.9.0
-scandir==1.6
-h5py==2.7.1
 Keras==2.2.4
 opencv-python==4.0.0.21
-tensorflow-gpu==1.13.1
+tensorflow-gpu==1.12.0
 plaidml-keras==0.5.0
 scikit-image
 dlib==19.16.0

requirements-opencl.txt

@@ -1,10 +1,8 @@
 numpy==1.16.1
-pathlib==1.0.1
+h5py==2.9.0
-scandir==1.6
-h5py==2.7.1
 Keras==2.2.4
 opencv-python==4.0.0.21
-tensorflow==1.13.1
+tensorflow==1.12.0
 plaidml-keras==0.5.0
 scikit-image
 tqdm

utils/Path_utils.py

@@ -1,5 +1,5 @@
 from pathlib import Path
-from scandir import scandir
+from os import scandir
 
 image_extensions = [".jpg", ".jpeg", ".png", ".tif", ".tiff"]