[MNT] add win action and resolve problem of HED under win

.github/workflows/build-and-test.yaml

@@ -11,7 +11,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-latest]
+        os: [ubuntu-latest, windows-latest]
         python-version: ['3.7', '3.8', '3.9', '3.10', '3.11']
     steps:
     - uses: actions/checkout@v2

docs/Examples/HED.rst

@@ -295,5 +295,5 @@ Perform pretraining, training and testing by invoking the ``pretrain``, ``train`
     weights_dir = osp.join(log_dir, "weights")
     
     bridge.pretrain("./weights")
-    bridge.train(train_data, val_data)
+    bridge.train(train_data, val_data, save_dir=weights_dir)
     bridge.test(test_data)

examples/hed/bridge.py

@@ -193,7 +193,7 @@ class HedBridge(SimpleBridge):
 
         return data_examples
 
-    def train(self, train_data, val_data, segment_size=10, min_len=5, max_len=8):
+    def train(self, train_data, val_data, segment_size=10, min_len=5, max_len=8, save_dir="./"):
         for equation_len in range(min_len, max_len):
             print_log(
                 f"============== equation_len: {equation_len}-{equation_len + 1} ================",
@@ -234,7 +234,9 @@ class HedBridge(SimpleBridge):
                     seems_good = self.check_rule_quality(rules, val_data, equation_len)
                     if seems_good:
                         self.reasoner.kb.learned_rules.update({equation_len: rules})
-                        self.model.save(save_path=f"./weights/eq_len_{equation_len}.pth")
+                        self.model.save(
+                            save_path=os.path.join(save_dir, f"eq_len_{equation_len}.pth")
+                        )
                         break
                     else:
                         if equation_len == min_len:
@@ -242,9 +244,13 @@ class HedBridge(SimpleBridge):
                                 "Learned mapping is: " + str(self.reasoner.idx_to_label),
                                 logger="current",
                             )
-                            self.model.load(load_path="./weights/pretrain_weights.pth")
+                            self.model.load(
+                                load_path=os.path.join(save_dir, f"pretrain_weights.pth")
+                            )
                         else:
-                            self.model.load(load_path=f"./weights/eq_len_{equation_len - 1}.pth")
+                            self.model.load(
+                                load_path=os.path.join(save_dir, f"eq_len_{equation_len - 1}.pth")
+                            )
                         condition_num = 0
                         print_log("Reload Model and retrain", logger="current")
 

examples/hed/datasets/get_dataset.py

@@ -4,8 +4,8 @@ import pickle
 import random
 import zipfile
 from collections import defaultdict
+from PIL import Image
 
-import cv2
 import gdown
 import numpy as np
 from torchvision.transforms import transforms
@@ -32,14 +32,15 @@ def get_pretrain_data(labels, image_size=(28, 28, 1)):
         label_path = osp.join(img_dir, label)
         img_path_list = os.listdir(label_path)
         for img_path in img_path_list:
-            img = cv2.imread(osp.join(label_path, img_path), cv2.IMREAD_GRAYSCALE)
-            img = cv2.resize(img, (image_size[1], image_size[0]))
-            X.append(np.array(img, dtype=np.float32))
+            with Image.open(osp.join(label_path, img_path)) as img:
+                img = img.convert('L')
+                img = img.resize((image_size[1], image_size[0]))
+                img_array = np.array(img, dtype=np.float32)
+                normalized_img = (img_array - 127) / 128.0
+                X.append(normalized_img)
 
-    X = [((img[:, :, np.newaxis] - 127) / 128.0) for img in X]
     Y = [img.copy().reshape(image_size[0] * image_size[1] * image_size[2]) for img in X]
-
-    X = [transform(img) for img in X]
+    X = [transform(img[:, :, np.newaxis]) for img in X]
     return X, Y
 
 

examples/hed/hed.ipynb

@@ -396,7 +396,7 @@
     "weights_dir = osp.join(log_dir, \"weights\")\n",
     "\n",
     "bridge.pretrain(weights_dir)\n",
-    "bridge.train(train_data, val_data)\n",
+    "bridge.train(train_data, val_data, save_dir=weights_dir)\n",
     "bridge.test(test_data)"
    ]
   }

examples/hed/main.py

@@ -95,7 +95,7 @@ def main():
     weights_dir = osp.join(log_dir, "weights")
 
     bridge.pretrain(weights_dir)
-    bridge.train(train_data, val_data)
+    bridge.train(train_data, val_data, save_dir=weights_dir)
     bridge.test(test_data)
 
 

examples/hed/reasoning/reasoning.py

@@ -10,6 +10,7 @@ CURRENT_DIR = os.path.abspath(os.path.dirname(__file__))
 
 class HedKB(PrologKB):
     def __init__(self, pseudo_label_list=[1, 0, "+", "="], pl_file=os.path.join(CURRENT_DIR, "learn_add.pl")):
+        pl_file = pl_file.replace("\\", "/")
         super().__init__(pseudo_label_list, pl_file)
         self.learned_rules = {}