Add PyTorch Project Convertor——CRAFT (#606)

* fix the code

* fix the visit_tuple

* Update stargan.md

* Update ultra_light_fast_generic_face_detector.md

* fix the docs

* remove static

* fix

* fix

* fix

* fix the docs

* fix the project_convertor

* remove

* fix nn.py

* modify the doc

* modify the doc

* fix
Co-authored-by: Nchanningss <chen_lingchi@163.com>

docs/pytorch_project_convertor/demo/README.md

@@ -5,5 +5,6 @@
 |------|-----|----------|------|
 | StaGAN | [demo](stargan.md)| [code](https://github.com/yunjey/stargan)|[code](https://github.com/SunAhong1993/stargan/tree/paddle)|
 | Ultra-Light-Fast-Generic-Face-Detector |[demo](ultra_light_fast_generic_face_detector.md)| [code](https://github.com/Linzaer/Ultra-Light-Fast-Generic-Face-Detector-1MB) |[code](https://github.com/SunAhong1993/Ultra-Light-Fast-Generic-Face-Detector-1MB/tree/paddle)|
+| CRAFT-pytorch |[demo](craft.md)| [code](https://github.com/clovaai/CRAFT-pytorch) |[code](https://github.com/SunAhong1993/CRAFT-pytorch/tree/paddle)|
 
 ***持续更新...***

x2paddle/project_convertor/pytorch/api_mapper/nn.py

@@ -386,10 +386,10 @@ class ReLUModuleMapper(Mapper):
                  target_name=None):
         super().__init__(func_name, pytorch_api_name, args, kwargs, target_name)
 
-    def delete_attrs(self):
-        delete_key(self.kwargs, "inplace")
-        if len(self.args) > 0:
-            self.args.clear()
+    def run(self):
+        if len(self.args) > 0 or len(self.kwargs) > 0:
+            self.func_name = "x2paddle.torch2paddle.ReLU"
+        return super().run()
 
 
 class SoftmaxModuleMapper(Mapper):

x2paddle/project_convertor/pytorch/api_mapper/ops.py

@@ -93,6 +93,8 @@ class SetDeviceMapper(Mapper):
     def run(self):
         self.process_attrs()
         insert_codes = list()
+        if self.target_name is None:
+            self.target_name = "tmp"
         insert_codes.append("{} = {}".format(self.target_name,
                                              self.useful_attrs["device"]))
         insert_codes.append("{} = {}.replace('cuda', 'gpu')".format(
@@ -539,19 +541,23 @@ class LinspaceMapper(Mapper):
                     out2, self.useful_attrs["requires_grad"])
             return out1, out2, out3
 
+
 class ToTensorMapper(Mapper):
     def __init__(self,
                  func_name,
                  pytorch_api_name,
-                 args, kwargs, 
+                 args,
+                 kwargs,
                  target_name=None):
         super().__init__(func_name, pytorch_api_name, args, kwargs, target_name)
 
     def process_attrs(self):
         rename_key(self.kwargs, "device", "place")
+
     def run(self):
         if self.rename_func_name("paddle.to_tensor"):
-            return [], generate_api_code(self.func_name, self.args, self.kwargs), []
+            return [], generate_api_code(self.func_name, self.args,
+                                         self.kwargs), []
         else:
             self.convert_args2kwargs()
             return self.convert_to_paddle()

x2paddle/project_convertor/pytorch/ast_update.py

@@ -497,6 +497,8 @@ class AstUpdater(ast.NodeVisitor):
         mapper = API_MAPPER[pytorch_api][1](func_name, pytorch_api, args_list,
                                             kw_dict, target_name)
         prefix_insert_codes, new_code, suffix_insert_codes = mapper.run()
+        if mapper.func_name.startswith("x2paddle."):
+            self.is_import_x2paddle = True
         scope_node = self._get_scope_node()
         if isinstance(ast.parse(new_code).body[0], ast.Assign):
             node_index = self._get_current_index(scope_node, node)

x2paddle/project_convertor/pytorch/models/resnet.py

@@ -79,7 +79,7 @@ class BasicBlock(nn.Layer):
         # Both self.conv1 and self.downsample layers downsample the input when stride != 1
         self.conv1 = conv3x3(inplanes, planes, stride)
         self.bn1 = norm_layer(planes)
-        self.relu = nn.ReLU()
+        self.relu = torch2paddle.ReLU(True)
         self.conv2 = conv3x3(planes, planes)
         self.bn2 = norm_layer(planes)
         self.downsample = downsample
@@ -133,7 +133,7 @@ class Bottleneck(nn.Layer):
         self.bn2 = norm_layer(width)
         self.conv3 = conv1x1(width, planes * self.expansion)
         self.bn3 = norm_layer(planes * self.expansion)
-        self.relu = nn.ReLU()
+        self.relu = torch2paddle.ReLU(True)
         self.downsample = downsample
         self.stride = stride
 
@@ -195,7 +195,7 @@ class ResNet(nn.Layer):
             padding=3,
             bias_attr=False)
         self.bn1 = norm_layer(self.inplanes)
-        self.relu = nn.ReLU()
+        self.relu = torch2paddle.ReLU(True)
         self.maxpool = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
         self.layer1 = self._make_layer(block, 64, layers[0])
         self.layer2 = self._make_layer(

x2paddle/project_convertor/pytorch/models/vgg.py

@@ -39,10 +39,10 @@ class VGG(nn.Layer):
         self.avgpool = nn.AdaptiveAvgPool2D((7, 7))
         self.classifier = nn.Sequential(
             nn.Linear(512 * 7 * 7, 4096),
-            nn.ReLU(),
+            torch2paddle.ReLU(True),
             nn.Dropout(),
             nn.Linear(4096, 4096),
-            nn.ReLU(),
+            torch2paddle.ReLU(True),
             nn.Dropout(),
             nn.Linear(4096, num_classes), )
         if init_weights:
@@ -81,9 +81,9 @@ def make_layers(cfg: List[Union[str, int]],
             v = cast(int, v)
             conv2d = nn.Conv2D(in_channels, v, kernel_size=3, padding=1)
             if batch_norm:
-                layers += [conv2d, nn.BatchNorm2D(v), nn.ReLU()]
+                layers += [conv2d, nn.BatchNorm2D(v), torch2paddle.ReLU(True)]
             else:
-                layers += [conv2d, nn.ReLU()]
+                layers += [conv2d, torch2paddle.ReLU(True)]
             in_channels = v
     return nn.Sequential(*layers)
 

x2paddle/project_convertor/pytorch/torch2paddle/nn.py

@@ -529,6 +529,19 @@ class MaxUnpool2D(paddle.nn.Layer):
         return out
 
 
+class ReLU(paddle.nn.ReLU):
+    def __init__(self, inplace=False):
+        super().__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        if self.inplace:
+            out = paddle.nn.functional.relu_(x)
+        else:
+            out = super().forward(x)
+        return out
+
+
 class ReflectionPad2D(paddle.nn.Pad2D):
     def __init__(self, padding):
         super().__init__(padding, mode="reflect")