update caffe2paddle

x2paddle/caffe_convert.py

+from x2paddle.decoder.caffe_decoder import CaffeDecoder
+Decoder = CaffeDecoder
+
+from x2paddle.op_mapper.dygraph.caffe2paddle.caffe_op_mapper import CaffeOpMapper
+DygraphOpMapper = CaffeOpMapper
+
+from x2paddle.op_mapper.static.caffe2paddle.caffe_op_mapper import CaffeOpMapper
+StaticOpMapper = CaffeOpMapper
+
+from x2paddle.optimizer.caffe_optimizer import CaffeOptimizer
+StaticOptimizer = CaffeOptimizer
\ No newline at end of file

x2paddle/convert.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 from six import text_type as _text_type
+from x2paddle import program
 import argparse
 import sys
 
@@ -66,8 +67,8 @@ def arg_parser():
     parser.add_argument(
         "--without_data_format_optimization",
         "-wo",
-        type=_text_type,
-        default="True",
+        action="store_true",
+        default=False,
         help="tf model conversion without data format optimization")
     parser.add_argument(
         "--define_input_shape",
@@ -88,11 +89,25 @@ def arg_parser():
         default=False,
         help="define whether merge the params")
     parser.add_argument(
-        "--input_shapes",
-        "-is",
+        "--jit_type",
+        "-jt",
+        type=_text_type,
+        default="script",
+        help="define the jit type of pytorch Module.")
+    parser.add_argument(
+        "--input_files",
+        "-if",
         action='append',
         default=None,
-        help="define the inputs' shape")
+        help="define the inputs' file path")
+    parser.add_argument(
+        "--paddle_type",
+        "-pt",
+        type=_text_type,
+        default="dygraph",
+        help="define the paddle model type after converting(dygraph/static)"
+    )
+    
     return parser
 
 
@@ -117,30 +132,27 @@ def tf2paddle(model_path,
             "[ERROR] Tensorflow is not installed, use \"pip install tensorflow\"."
         )
         return
-    from x2paddle import program
+
     from x2paddle.decoder.tf_decoder import TFDecoder
     from x2paddle.op_mapper.tf_op_mapper import TFOpMapper
-    from x2paddle.optimizer.tensorflow.bias import BiasOpt
-    from x2paddle.optimizer.tensorflow.transpose import TransposeOpt
-    from x2paddle.optimizer.tensorflow.batch_norm import BatchNormOpt
+    from x2paddle.op_mapper.tf_op_mapper_nhwc import TFOpMapperNHWC
+    from x2paddle.optimizer.tf_optimizer import TFOptimizer
 
     print("Now translating model from tensorflow to paddle.")
     model = TFDecoder(model_path, define_input_shape=define_input_shape)
-    mapper = TFOpMapper(model)
+
+    mapper = TFOpMapperNHWC(model)
     program.build()
-    bias_opt = BiasOpt()
-    transpose_opt = TransposeOpt()
-    batch_norm_opt = BatchNormOpt()
-    bias_opt.run(program)
-    batch_norm_opt.run(program)
-    transpose_opt.run(program)
     program.gen_model(save_dir)
 
 
-def caffe2paddle(proto, weight, save_dir, caffe_proto, params_merge=False):
-    from x2paddle.decoder.caffe_decoder import CaffeDecoder
-    from x2paddle.op_mapper.caffe_op_mapper import CaffeOpMapper
-    from x2paddle.optimizer.caffe_optimizer import CaffeOptimizer
+def caffe2paddle(proto, weight, save_dir, caffe_proto, 
+                 paddle_type, params_merge=False):
+    from x2paddle.caffe_convert import Decoder
+    if paddle_type == "dygraph":
+        from x2paddle.caffe_convert import DygraphOpMapper as OpMapper
+    else:
+        from x2paddle.caffe_convert import StaticOpMapper as OpMapper
     import google.protobuf as gpb
     ver_part = gpb.__version__.split('.')
     version_satisfy = False
@@ -149,12 +161,10 @@ def caffe2paddle(proto, weight, save_dir, caffe_proto, params_merge=False):
         version_satisfy = True
     assert version_satisfy, '[ERROR] google.protobuf >= 3.6.0 is required'
     print("Now translating model from caffe to paddle.")
-    model = CaffeDecoder(proto, weight, caffe_proto)
-    mapper = CaffeOpMapper(model)
-    optimizer = CaffeOptimizer(mapper)
-    optimizer.merge_bn_scale()
-    optimizer.merge_op_activation()
-    mapper.save_inference_model(save_dir, params_merge)
+    model = Decoder(proto, weight, caffe_proto)
+    mapper = OpMapper(model)
+    mapper.pd_graph.build()
+    mapper.pd_graph.gen_model(save_dir)
 
 
 def onnx2paddle(model_path, save_dir, params_merge=False):
@@ -162,8 +172,8 @@ def onnx2paddle(model_path, save_dir, params_merge=False):
     try:
         import onnx
         version = onnx.version.version
-        if version < '1.6.0':
-            print("[ERROR] onnx>=1.6.0 is required")
+        if version != '1.6.0':
+            print("[ERROR] onnx==1.6.0 is required")
             return
     except:
         print("[ERROR] onnx is not installed, use \"pip install onnx==1.6.0\".")
@@ -185,7 +195,7 @@ def onnx2paddle(model_path, save_dir, params_merge=False):
     print("Paddle model and code generated.")
 
 
-def pytorch2paddle(model_path, save_dir, input_shapes):
+def pytorch2paddle(model_path, save_dir, jit_type, input_files):
     # check pytorch installation and version
     try:
         import torch
@@ -202,9 +212,12 @@ def pytorch2paddle(model_path, save_dir, input_shapes):
         return
     print("Now translating model from pytorch to paddle.")
 
-    from x2paddle.decoder.pytorch_decoder import PyTorchDecoder
+    from x2paddle.decoder.pytorch_decoder import ScriptDecoder, TraceDecoder
     from x2paddle.op_mapper.pytorch2paddle import pytorch_op_mapper
-    model = PyTorchDecoder(model_path)
+    if jit_type == "trace":
+        model = TraceDecoder(model_path, input_files)
+    else:
+        model = ScriptDecoder(model_path)
     mapper = pytorch_op_mapper.PyTorchOpMapper(model)
     mapper.graph.build()
     print("Model optimizing ...")
@@ -212,34 +225,15 @@ def pytorch2paddle(model_path, save_dir, input_shapes):
     graph_opt = GraphOptimizer()
     graph_opt.optimize(mapper.graph)
     print("Model optimized.")
-    if input_shapes is not None:
-        real_input_shapes = list()
-        for shape in input_shapes:
-            sp = shape[1:-1].split(",")
-            for i, s in enumerate(sp):
-                sp[i] = int(s)
-            real_input_shapes.append(sp)
-    else:
-        real_input_shapes = None
-    mapper.graph.gen_model(save_dir, real_input_shapes)
+    mapper.graph.gen_model(save_dir, jit_type, input_files)
 
 
 def paddle2onnx(model_path, save_dir, opset_version=10):
-    import paddle.fluid as fluid
-    try:
-        import paddle2onnx
-    except:
-        print(
-            "[ERROR] paddle2onnx not installed, use \"pip install paddle2onnx\"")
-
-    import paddle2onnx as p2o
-    model = p2o.PaddleDecoder(model_path, '__model__', '__params__')
-    mapper = p2o.PaddleOpMapper()
-    mapper.convert(
-        model.program,
-        save_dir,
-        scope=fluid.global_scope(),
-        opset_version=opset_version)
+    from x2paddle.decoder.paddle_decoder import PaddleDecoder
+    from x2paddle.op_mapper.paddle2onnx.paddle_op_mapper import PaddleOpMapper
+    model = PaddleDecoder(model_path, '__model__', '__params__')
+    mapper = PaddleOpMapper()
+    mapper.convert(model.program, save_dir, opset_number=opset_version)
 
 
 def main():
@@ -260,6 +254,7 @@ def main():
 
     assert args.framework is not None, "--framework is not defined(support tensorflow/caffe/onnx)"
     assert args.save_dir is not None, "--save_dir is not defined"
+    assert args.paddle_type in ["dygraph", "static"], "--paddle_type must be 'dygraph' or 'static'"
 
     try:
         import paddle
@@ -267,8 +262,8 @@ def main():
         print("paddle.__version__ = {}".format(paddle.__version__))
         if v0 == '0' and v1 == '0' and v2 == '0':
             print("[WARNING] You are use develop version of paddlepaddle")
-        elif int(v0) != 1 or int(v1) < 6:
-            print("[ERROR] paddlepaddle>=1.6.0 is required")
+        elif int(v0) != 2 or int(v1) < 0:
+            print("[ERROR] paddlepaddle>=2.0.0 is required")
             return
     except:
         print(
@@ -277,12 +272,11 @@ def main():
 
     if args.framework == "tensorflow":
         assert args.model is not None, "--model should be defined while translating tensorflow model"
-        assert args.without_data_format_optimization in [
-            "True", "False"
-        ], "--the param without_data_format_optimization should be defined True or False"
+        without_data_format_optimization = False
         define_input_shape = False
         params_merge = False
-        without_data_format_optimization = True if args.without_data_format_optimization == "True" else False
+        if args.without_data_format_optimization:
+            without_data_format_optimization = True
         if args.define_input_shape:
             define_input_shape = True
         if args.params_merge:
@@ -296,7 +290,7 @@ def main():
         if args.params_merge:
             params_merge = True
         caffe2paddle(args.prototxt, args.weight, args.save_dir,
-                     args.caffe_proto, params_merge)
+                     args.caffe_proto, args.paddle_type, params_merge)
     elif args.framework == "onnx":
         assert args.model is not None, "--model should be defined while translating onnx model"
         params_merge = False
@@ -304,10 +298,13 @@ def main():
         if args.params_merge:
             params_merge = True
         onnx2paddle(args.model, args.save_dir, params_merge)
-        
+    elif args.framework == "pytorch":
+        assert args.model is not None, "--model should be defined while translating pytorch model"
+        pytorch2paddle(args.model, args.save_dir, args.jit_type, args.input_files)
+
     elif args.framework == "paddle2onnx":
         assert args.model is not None, "--model should be defined while translating paddle model to onnx"
-        paddle2onnx(args.model, args.save_dir, opset_version=args.onnx_opset)
+        paddle2onnx(args.model, args.save_dir, args.onnx_opset)
 
     else:
         raise Exception(

x2paddle/core/program.py

-# -*- coding:UTF-8 -*-
 #   Copyright (c) 2020  PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License"
@@ -21,15 +20,15 @@ import paddle
 from paddle.fluid.proto import framework_pb2
 from collections import OrderedDict
 import numpy
-import collections
 import sys
 import os
 import six
 import pickle
+import numpy as np
 
 
 class PaddleLayer(object):
-    def __init__(self, id, kernel, inputs, outputs, **kwargs):
+    def __init__(self, id, kernel, inputs, outputs, scope_name="", **kwargs):
         assert isinstance(
             inputs,
             dict), "parameter 'inputs' for PaddleLayer should be type of dict"
@@ -53,9 +52,11 @@ class PaddleLayer(object):
         self.kernel = kernel
         self.inputs = inputs
         self.outputs = outputs
+        self.scope_name = scope_name
         self.attrs = kwargs
         self.id = id
         self.blocks = list()
+        
 
     def add_block(self, block):
         self.blocks.append(block)
@@ -71,12 +72,23 @@ class PaddleGraph(object):
         self.parameters = dict()
         self.parent_layer = parent_layer
         self.graph_type = graph_type
+        self.custom_func = None
+        self.inputs_info = None
 
     def set_name(self, name):
-        self.name = name
+        self.name = name.replace("-", "_").replace("/", "_")
 
     def set_parameters(self, parameters):
         self.parameters = parameters
+        
+    def set_custom_func(self, custom_func):
+        self.custom_func = custom_func
+        
+    def set_inputs_info(self, inputs_info):
+        self.inputs_info = inputs_info
+        
+    def set_script(self, script):
+        self.script = script
 
     def clear(self):
         self.layers = OrderedDict()
@@ -90,13 +102,13 @@ class PaddleGraph(object):
         self.edges_out = dict()
         self.edges_in = dict()
 
-    def add_layer(self, kernel, inputs, outputs, **kwargs):
+    def add_layer(self, kernel, inputs, outputs, scope_name="", **kwargs):
         layer_id = str(len(self.layers))
         if self.parent_layer is not None:
             layer_id = "{}.{}.{}".format(self.parent_layer.id,
                                          len(self.parent_layer.blocks),
                                          layer_id)
-        layer = PaddleLayer(layer_id, kernel, inputs, outputs, **kwargs)
+        layer = PaddleLayer(layer_id, kernel, inputs, outputs, scope_name=scope_name, **kwargs)
         self.layers[layer_id] = layer
         return layer_id
 
@@ -156,6 +168,10 @@ class PaddleGraph(object):
                 if not isinstance(vs, list):
                     vs = [vs]
                 for v in vs:
+                    if "[" in v:
+                        remove_index = v.index("[")
+                        v_part = v[remove_index:]
+                        v = v.replace(v_part, "")
                     assert v in outputs_from_nodes or (
                         inputs is not None and v in list(inputs.values())
                     ) or (
@@ -215,8 +231,59 @@ class PaddleGraph(object):
                     block_global_layers = update(block.layers)
                     global_layers.update(block_global_layers)
             return global_layers
-
         return update(self.layers)
+    
+    def gen_model(self, save_dir, jit_type=None):
+        if not os.path.exists(save_dir):
+            os.makedirs(save_dir)
+        if self.graph_type == "static":
+            code_dir = os.path.join(save_dir, 'model_with_code')
+            infer_dir = os.path.join(save_dir, 'inference_model')
+            self.gen_code(code_dir)
+            sys.path.append(code_dir)
+            import x2paddle_model
+            paddle.enable_static()
+            scope = paddle.static.Scope()
+            startup_program = paddle.static.Program()
+            main_program = paddle.static.Program()
+            with paddle.static.scope_guard(scope):
+                with paddle.static.program_guard(main_program, startup_program):
+                    inputs, outputs = x2paddle_model.x2paddle_net()
+                    exe = fluid.Executor(fluid.CPUPlace())
+                    exe.run(startup_program)
+                    param_dir = os.path.join(code_dir, 'weights')
+                    for k, v in self.parameters.items():
+                        if scope.find_var(k):
+                            self.dump_parameter(k, v, param_dir)
+                    def if_exist(var):
+                        b = os.path.exists(
+                            os.path.join(os.path.join(param_dir, var.name)))
+                        return b
+                    fluid.io.load_vars(
+                        exe, param_dir, main_program, predicate=if_exist)
+                    fluid.io.save_inference_model(
+                        dirname=infer_dir,
+                        feeded_var_names=[i.name for i in inputs],
+                        target_vars=outputs,
+                        executor=exe)
+        else:
+            if jit_type == "trace":
+                from x2paddle.optimizer.code_optimizer import HierarchicalTree
+                hierarchical_tree = HierarchicalTree(self)
+                for layer_id, layer in self.layers.items():
+                    hierarchical_tree.insert(layer)
+                hierarchical_tree.save_source_files(save_dir)
+                self.dump_dygraph_parameter(save_dir)
+            else:
+                self.gen_dygraph_code(save_dir)
+                self.dump_dygraph_parameter(save_dir)
+            input_shapes = list()
+            input_types = list()
+            for input_name in self.inputs:
+                input_shapes.append(self.inputs_info[input_name][0])
+                input_types.append(self.inputs_info[input_name][1])
+                # 如果input_files非空，则导出推理模型；其值类似[[None, 3, 224, 224]]
+                self.dygraph2static(save_dir, input_shapes, input_types)
 
     def gen_code(self, code_dir):
         def write_code(f, code_list, indent=0):
@@ -235,10 +302,24 @@ class PaddleGraph(object):
             f, [
                 "from paddle.fluid.initializer import Constant",
                 "from paddle.fluid.param_attr import ParamAttr",
-                "import paddle.fluid as fluid", "import math", "",
-                "def x2paddle_net():"
+                "import paddle.fluid as fluid", 
+                "import paddle", "import math", "",
+                
             ],
             indent=0)
+        if self.custom_func is not None:
+            write_code(
+                f, 
+                list(self.custom_func.values()),
+                indent=0)
+        write_code(f, 
+            ["", "def x2paddle_net():"],
+            indent=0)
+        write_code(
+            f, [
+                "paddle.enable_static()"
+            ],
+            indent=1)
         for layer_id, layer in self.layers.items():
             edges_in = self.edges_in.get(layer_id, [])
             edges_out = self.edges_out.get(layer_id, [])
@@ -253,8 +334,10 @@ class PaddleGraph(object):
                 for output in layer.outputs:
                     line += "{}, ".format(output)
                 line = line.strip(", ")
-
-            line += " = {}(".format(layer.kernel)
+            if layer.kernel.startswith("combination_layer"):
+                line += " = {}(".format(layer.kernel.split(":")[-1].lower() + "_layer")
+            else:
+                line += " = {}(".format(layer.kernel)
             for k, v in layer.inputs.items():
                 if isinstance(v, list):
                     line += "{}=[{}], ".format(k, ", ".join(v))
@@ -274,47 +357,6 @@ class PaddleGraph(object):
             indent=1)
         f.close()
 
-    def gen_model(self, save_dir, input_shapes=None):
-        if not os.path.exists(save_dir):
-            os.makedirs(save_dir)
-        if self.graph_type == "static":
-            code_dir = os.path.join(save_dir, 'model_with_code')
-            infer_dir = os.path.join(save_dir, 'inference_model')
-            self.gen_code(code_dir)
-            sys.path.append(code_dir)
-            import x2paddle_model
-            scope = fluid.Scope()
-            startup_program = fluid.Program()
-            main_program = fluid.Program()
-            with fluid.scope_guard(scope):
-                with fluid.program_guard(main_program, startup_program):
-                    inputs, outputs = x2paddle_model.x2paddle_net()
-                    exe = fluid.Executor(fluid.CPUPlace())
-                    exe.run(startup_program)
-
-                    param_dir = os.path.join(code_dir, 'weights')
-                    for k, v in self.parameters.items():
-                        if scope.find_var(k):
-                            self.dump_parameter(k, v, param_dir)
-
-                    def if_exist(var):
-                        b = os.path.exists(
-                            os.path.join(os.path.join(param_dir, var.name)))
-                        return b
-
-                    fluid.io.load_vars(
-                        exe, param_dir, main_program, predicate=if_exist)
-                    fluid.io.save_inference_model(
-                        dirname=infer_dir,
-                        feeded_var_names=[i.name for i in inputs],
-                        target_vars=outputs,
-                        executor=exe)
-        else:
-            self.gen_dygraph_code(save_dir)
-            self.dump_dygraph_parameter(save_dir)
-            if input_shapes is not None:
-                # 如果input_shapes非空，则导出推理模型；其值类似[[None, 3, 224, 224]]
-                self.dygraph2static(save_dir, input_shapes)
 
     def dump_parameter(self, param_name, param, save_dir):
         if not os.path.exists(save_dir):
@@ -356,10 +398,10 @@ class PaddleGraph(object):
                 if self.edges_in.get(layer_id, 0) == 0 and self.edges_out.get(
                         layer_id, 0) == 0:
                     continue
-                if layer.kernel == "fluid.dygraph.base.to_variable":
-                    value = layer.attrs["value"]
-                    if not value.startswith("params["):
-                        self.inputs.append(value)
+                if layer.kernel == "paddle.to_tensor":
+                    data = layer.attrs["data"]
+                    if not data.startswith("params["):
+                        self.inputs.append(data)
                 if len(layer.blocks) > 0:
                     for block in layer.blocks:
                         block.get_dygraph_inputs()
@@ -376,11 +418,15 @@ class PaddleGraph(object):
                     layer_id, 0) == 0:
                 continue
             if self.edges_out.get(layer_id, 0) == 0:
-                for output_name in layer.outputs:
-                    if not output_name.startswith("x"):
-                        continue
-                    self.outputs.append(output_name)
-        self.outputs = list(set(self.outputs))
+                
+                for i, output_name in enumerate(layer.outputs):
+                    if ("paddle.nn" in layer.kernel and "functional" not in layer.kernel) or \
+                            (layer.kernel == "paddle.to_tensor" and layer.attrs["data"].startswith("params["))or \
+                            "paddle.fluid.dygraph" in layer.kernel:
+                        if i == 0:
+                            continue
+                    if output_name not in self.outputs:
+                        self.outputs.append(output_name)
 
     def gen_dygraph_code(self, code_dir=None, indent=2):
         def gen_codes(code_list, indent=0):
@@ -415,6 +461,23 @@ class PaddleGraph(object):
                 gen_codes(
                     ["def forward(self, {}):".format(input_data_name)],
                     indent=1))
+            
+        def gen_run_net_code(code_dir):
+            input_data_name = ', '.join(self.inputs)
+            self.run_func = gen_codes(
+                [
+                    "",
+                    "def run_net({}):".format(input_data_name),
+                ],
+                indent=0)
+            self.run_func.extend(
+                gen_codes(["paddle.disable_static()",
+                           "params, _ = fluid.load_dygraph('{}/model')".format(code_dir),
+                           "model = {}(params)".format(self.name),
+                           "model.set_dict(params)",
+                           "model.eval()",
+                           "out = model({})".format(input_data_name),
+                           "return out"], indent=1))
 
         def write_code(code_dir):
             f = open(os.path.join(code_dir, 'x2paddle_code.py'), 'w')
@@ -431,6 +494,8 @@ class PaddleGraph(object):
             self.forward_func.extend(gen_codes([return_code], indent=2))
             for code_line in self.forward_func:
                 f.write(code_line)
+            for code_line in self.run_func:
+                f.write(code_line)
             f.close()
 
         self.init_func = []
@@ -440,12 +505,12 @@ class PaddleGraph(object):
 
         for layer_id, layer in self.layers.items():
             if ("paddle.nn" in layer.kernel and "functional" not in layer.kernel
-                ) or layer.kernel == "fluid.dygraph.base.to_variable" or \
+                ) or layer.kernel == "paddle.to_tensor" or \
                "paddle.fluid.dygraph" in layer.kernel:
                 line = "{}".format(
                     layer.outputs[0]
-                ) if layer.kernel == "fluid.dygraph.base.to_variable" and not layer.attrs[
-                    "value"].startswith("params[") else "self.{}".format(
+                ) if layer.kernel == "paddle.to_tensor" and not layer.attrs[
+                    "data"].startswith("params[") else "self.{}".format(
                         layer.outputs[0])
                 line += " = {}(".format(layer.kernel)
                 for k, v in layer.attrs.items():
@@ -453,8 +518,8 @@ class PaddleGraph(object):
                 line = line.strip(", ")
                 line += ")"
 
-                if layer.kernel == "fluid.dygraph.base.to_variable" and not layer.attrs[
-                        "value"].startswith("params["):
+                if layer.kernel == "paddle.to_tensor" and not layer.attrs[
+                        "data"].startswith("params["):
                     self.forward_func.extend(gen_codes([line], indent=indent))
                     continue
                 else:
@@ -466,8 +531,8 @@ class PaddleGraph(object):
                     line = layer.outputs[1]
                 else:
                     line = ','.join(layer.outputs[1:])
-                if layer.kernel == "fluid.dygraph.base.to_variable" and layer.attrs[
-                        "value"].startswith("params["):
+                if layer.kernel == "paddle.to_tensor" and layer.attrs[
+                        "data"].startswith("params["):
                     line += " = self.{}".format(layer.outputs[0])
                 else:
                     line += " = self.{}(".format(layer.outputs[0])
@@ -478,7 +543,7 @@ class PaddleGraph(object):
                 self.forward_func.extend(gen_codes([line], indent=indent))
             elif "prim" in layer.kernel:
                 func_name = layer.kernel.replace(".", "_")
-                from x2paddle.op_mapper.pytorch2paddle import prim2code
+                from x2paddle.op_mapper.dygraph import prim2code
                 if hasattr(prim2code, func_name):
                     func = getattr(prim2code, func_name)
                     func(
@@ -504,6 +569,7 @@ class PaddleGraph(object):
                 line += ")"
                 self.forward_func.extend(gen_codes([line], indent=indent))
         if indent == 2:
+            gen_run_net_code(code_dir)
             write_code(code_dir)
         else:
             return self.init_func, self.forward_func
@@ -513,23 +579,22 @@ class PaddleGraph(object):
         pickle.dump(self.parameters, params_output)
         params_output.close()
 
-    def dygraph2static(self, save_dir, input_shapes=[]):
+    def dygraph2static(self, save_dir, input_shapes=[], input_types=[]):
         from paddle.fluid.dygraph.jit import declarative
         sepc_list = list()
         for i, name in enumerate(self.inputs):
+            input_shapes[i][0] = -1
             sepc_list.append(
                 paddle.static.InputSpec(
-                    shape=input_shapes[i], name=name))
+                    shape=input_shapes[i], name=name, dtype=input_types[i]))
         import sys
         path = osp.abspath(save_dir)
         sys.path.insert(0, save_dir)
         import x2paddle_code
-        place = fluid.CPUPlace()
-        with fluid.dygraph.guard(place):
-            restore, _ = fluid.load_dygraph(osp.join(save_dir, "model"))
-            model = getattr(x2paddle_code, self.name)(restore)
-            model.set_dict(restore)
-            model.eval()
-            model.forward = declarative(model.forward, sepc_list)
-        fluid.dygraph.jit.save(
-            layer=model, model_path=osp.join(save_dir, "inference"))
+        paddle.disable_static()
+        restore, _ = fluid.load_dygraph(osp.join(save_dir, "model"))
+        model = getattr(x2paddle_code, self.name)(restore)
+        model.set_dict(restore)
+        model.eval()
+        static_model = paddle.jit.to_static(model, input_spec=sepc_list)
+        paddle.jit.save(static_model, osp.join(save_dir, "inference_model/model"))
\ No newline at end of file

x2paddle/decoder/caffe_decoder.py

@@ -18,7 +18,6 @@ from google.protobuf import text_format
 import numpy as np
 from x2paddle.core.graph import GraphNode, Graph
 from x2paddle.core.fluid_code import FluidCode
-from x2paddle.op_mapper import caffe_shape
 
 
 class CaffeResolver(object):
@@ -50,10 +49,10 @@ class CaffeGraphNode(GraphNode):
     def __init__(self, layer, type_str, layer_name=None):
         if layer_name is None:
             super(CaffeGraphNode, self).__init__(
-                layer, layer.name.replace('/', '_').replace('-', '_'))
+                layer, layer.name.replace('/', '_').replace('-', '_').lower())
         else:
             super(CaffeGraphNode, self).__init__(
-                layer, layer_name.replace('/', '_').replace('-', '_'))
+                layer, layer_name.replace('/', '_').replace('-', '_').lower())
         self.layer_type = type_str
         self.fluid_code = FluidCode()
         self.data = None
@@ -66,6 +65,13 @@ class CaffeGraph(Graph):
     def __init__(self, model, params, caffe_pb):
         self.params = params
         self.caffe_pb = caffe_pb
+        if hasattr(model, "name"):
+            if model.name == "":
+                self.graph_name = "CaffeModel"
+            else:
+                self.graph_name = model.name
+        else:
+            self.graph_name = "CaffeModel"
         super(CaffeGraph, self).__init__(model)
 
     def filter_layers(self, layers):
@@ -242,7 +248,7 @@ class CaffeDecoder(object):
         with open(proto_path, 'rb') as proto_file:
             proto_str = proto_file.read()
             text_format.Merge(proto_str, self.net)
-
+        
         self.load_using_pb()
 
         self.caffe_graph = CaffeGraph(self.net, self.params,

x2paddle/decoder/pytorch_decoder.py

@@ -12,15 +12,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import os
+import sys
 import torch
+import numpy as np
 
 
-class PyTorchDecoder(object):
-    def __init__(self, script_path):
-        self.script = torch.jit.load(script_path)
-        self.graph = self._optimize_graph(self.script.inlined_graph)
-
-    def _optimize_graph(self, graph):
+class Decoder(object):
+     def _optimize_graph(self, graph):
         torch._C._jit_pass_constant_propagation(graph)
         torch._C._jit_pass_dce(graph)
         torch._C._jit_pass_lint(graph)
@@ -31,4 +30,37 @@ class PyTorchDecoder(object):
         torch._C._jit_pass_canonicalize(graph)
         torch._C._jit_pass_lint(graph)
         torch._C._jit_pass_constant_propagation(graph)
-        return graph
+        return graph       
+
+
+class ScriptDecoder(Decoder):
+    """ 当script_path非None，直接load ScriptModule;
+        当model_path非None，load PyTorchModule后使用script方式转换为ScriptModule。
+        
+        Args:
+            script_path (str): ScriptModule保存路径。
+            model_path (str): PyTorchModule保存路径。
+    """
+    def __init__(self, script_path=None):
+        self.script = torch.jit.load(script_path)
+        self.graph = self._optimize_graph(self.script.inlined_graph)
+            
+class TraceDecoder(Decoder):
+    """ PyTorchModule后使用trace方式转换为ScriptModule。
+        
+        Args:
+            model_path (str): PyTorchModule保存路径。
+            input_files (list): 输入网络的numpy，每个numpy保存成.npy文件, 
+                                文件路径存储在input_files中。
+    """
+    def __init__(self, model_path, input_files=list()):
+        # TODO(syf): 传入pytorch的Module(即import)，否则出错
+        model = torch.load(model_path)
+        model.eval()
+        input_list = list()
+        for npy_file in input_files:
+            input_list.append(torch.tensor(np.load(npy_file)))
+        self.script = torch.jit.trace(model, input_list, strict=False)
+        self.graph = self._optimize_graph(self.script.inlined_graph)
+#         print(self.graph)
+#         print(getattr(getattr(self.script.decoder.block, "5").layer, "2"))

x2paddle/op_mapper/dygraph/caffe2paddle/caffe_shape.py

+#   Copyright (c) 2019  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import numbers
+from functools import reduce
+
+
+def get_kernel_parameters(params):
+    [k_h, k_w] = [1, 1]
+    if isinstance(params.kernel_size, numbers.Number):
+        [k_h, k_w] = [params.kernel_size] * 2
+    elif len(params.kernel_size) > 0:
+        k_h = params.kernel_h if params.kernel_h > 0 else params.kernel_size[0]
+        k_w = params.kernel_w if params.kernel_w > 0 else params.kernel_size[
+            len(params.kernel_size) - 1]
+    elif params.kernel_h > 0 or params.kernel_w > 0:
+        k_h = params.kernel_h
+        k_w = params.kernel_w
+    [s_h, s_w] = [1, 1]
+    if isinstance(params.stride, numbers.Number):
+        [s_h, s_w] = [params.stride] * 2
+    elif len(params.stride) > 0:
+        s_h = params.stride_h if params.stride_h > 0 else params.stride[0]
+        s_w = params.stride_w if params.stride_w > 0 else params.stride[len(
+            params.stride) - 1]
+    elif params.stride_h > 0 or params.stride_w > 0:
+        s_h = params.stride_h
+        s_w = params.stride_w
+    [p_h, p_w] = [0, 0]
+    if isinstance(params.pad, numbers.Number):
+        [p_h, p_w] = [params.pad] * 2
+    elif len(params.pad) > 0:
+        p_h = params.pad_h if params.pad_h > 0 else params.pad[0]
+        p_w = params.pad_w if params.pad_w > 0 else params.pad[len(params.pad) -
+                                                               1]
+    elif params.pad_h > 0 or params.pad_w > 0:
+        p_h = params.pad_h
+        p_w = params.pad_w
+    dila_h = dila_w = 1
+    if hasattr(params, 'dilation'):
+        dila_len = len(params.dilation)
+        if dila_len == 2:
+            dila_h = params.dilation[0]
+            dila_w = params.dilation[1]
+        elif dila_len == 1:
+            dila_h = dila_w = params.dilation[0]
+        else:
+            assert dila_len == 0, "invalid length[%s] of dilation in convolution" % (
+                dila_len)
+    return dila_h, dila_w, p_h, p_w, k_h, k_w, s_h, s_w
+
+
+def get_strided_kernel_output_shape(params, input_shape, round_func):
+    i_h = input_shape[2]
+    i_w = input_shape[3]
+    dila_h, dila_w, pad_h, pad_w, kernel_h, kernel_w, stride_h, stride_w = get_kernel_parameters(
+        params)
+    o_h = (i_h + 2 * pad_h - (dila_h *
+                              (kernel_h - 1) + 1)) / float(stride_h) + 1
+    o_w = (i_w + 2 * pad_w - (dila_w *
+                              (kernel_w - 1) + 1)) / float(stride_w) + 1
+    o_h = int(round_func(o_h))
+    o_w = int(round_func(o_w))
+    has_c_o = hasattr(params, 'num_output')
+    c = params.num_output if has_c_o else input_shape[1]
+    return [[input_shape[0], c, o_h, o_w]]
+
+
+def shape_convolution(layer, input_shape):
+    params = layer.convolution_param
+    return get_strided_kernel_output_shape(params, input_shape[0], math.floor)
+
+
+def shape_deconvolution(layer, input_shape):
+
+    h_i = input_shape[0][2]
+    w_i = input_shape[0][3]
+
+    params = layer.convolution_param
+    dila_h, dila_w, pad_h, pad_w, kernel_h, kernel_w, stride_h, stride_w = get_kernel_parameters(
+        params)
+
+    h_o = (h_i - 1) * stride_h - 2 * pad_h + dila_h * (kernel_h - 1) + 1
+    w_o = (w_i - 1) * stride_w - 2 * pad_w + dila_w * (kernel_w - 1) + 1
+
+    has_c_o = hasattr(params, 'num_output')
+    c = params.num_output if has_c_o else input_shape.channels
+    return [[input_shape[0][0], c, h_o, w_o]]
+
+
+def shape_pooling(layer, input_shape):
+    params = layer.pooling_param
+    global_pool = getattr(params, 'global_pooling', False)
+    if global_pool:
+        return [[input_shape[0][0], input_shape[0][1], 1, 1]]
+
+    ceil_mode = getattr(params, 'ceil_mode', True)
+    if ceil_mode is True:
+        method = math.ceil
+    else:
+        method = math.floor
+    return get_strided_kernel_output_shape(params, input_shape[0], method)
+
+
+def shape_convolutiondepthwise(layer, input_shape):
+    params = layer.convolution_param
+    return get_strided_kernel_output_shape(params, input_shape[0], math.floor)
+
+
+def shape_innerproduct(layer, input_shape):
+    params = layer.inner_product_param
+    return [[input_shape[0][0], params.num_output]]
+
+
+def shape_lrn(layer, input_shape):
+    return input_shape
+
+
+def shape_relu(layer, input_shape):
+    return input_shape
+
+
+def shape_softmax(layer, input_shape):
+    return input_shape
+
+
+def shape_input(layer, input_shape):
+    return [list(layer.input_param.shape[0].dim)]
+
+
+def shape_memorydata(layer, input_shape):
+    params = layer.memory_data_param
+    shape = []
+    shape.append(int(params.batch_size))
+    shape.append(int(params.channels))
+    shape.append(int(params.height))
+    shape.append(int(params.width))
+    return [shape]
+
+
+def shape_concat(layer, input_shape):
+    params = layer.concat_param
+    axis = params.axis
+    output_shape = None
+    for shape in input_shape:
+        if output_shape is None:
+            output_shape = []
+            for i in range(len(shape)):
+                output_shape.append(shape[i])
+        else:
+            output_shape[axis] += shape[axis]
+    return [output_shape]
+
+
+def shape_slice(layer, input_shape):
+    inshape = input_shape[0]
+
+    top_len = len(layer.top)
+    params = layer.slice_param
+    axis = params.axis
+    slice_dim = params.slice_dim
+    if slice_dim != 1 and axis == 1:
+        axis = slice_dim
+    points = list(params.slice_point)
+    count = inshape[axis]
+    if len(points) == 0:
+        assert count % top_len == 0, "the parameter of Slice is wrong"
+        part = count / top_len
+        t = part
+        while t < count:
+            points.append(int(t))
+            t += part
+    points = [0] + points + [count]
+    output_shape = []
+    for i in range(len(points)):
+        shape = []
+        for ii in range(len(inshape)):
+            shape.append(inshape[ii])
+        size = points[i + 1] - points[i]
+        shape[axis] = size
+        output_shape.append(shape)
+        if i == len(points) - 2:
+            break
+    return output_shape
+
+
+def shape_prelu(layer, input_shape):
+    return input_shape
+
+
+def shape_sigmoid(layer, input_shape):
+    return input_shape
+
+
+def shape_absval(layer, input_shape):
+    return input_shape
+
+
+def shape_accuracy(layer, input_shape):
+    return [[1]]
+
+
+def shape_tanh(layer, input_shape):
+    return input_shape
+
+
+def shape_eltwise(layer, input_shape):
+    return [input_shape[0]]
+
+
+def shape_batchnorm(layer, input_shape):
+    return input_shape
+
+
+def shape_scale(layer, input_shape):
+    return input_shape
+
+
+def shape_reshape(layer, input_shape):
+    def count(num_list):
+        return reduce(lambda a, b: a * b, num_list)
+
+    inshape = input_shape[0]
+    params = layer.reshape_param
+    axis = params.axis if hasattr(params, 'axis') else 0
+    num_axes = params.num_axes if hasattr(params, 'num_axes') else -1
+    if inshape[0] == -1:
+        inshape[0] = 1
+    input_count = count(inshape)
+
+    input_num_axes = len(inshape)
+
+    input_start_axis = axis
+    start_axis = input_start_axis if input_start_axis >= 0 \
+            else input_num_axes + input_start_axis + 1
+
+    assert start_axis >= 0, "[Reshape]axis %d out of range" % (input_start_axis)
+    assert start_axis <= input_num_axes, "[Reshape]axis %d out of range for %d-D input data"\
+            % (input_start_axis, input_num_axes)
+
+    assert num_axes >= -1, "[Reshape]num_axes must be >= 0, or -1 for all"
+
+    end_axis = input_num_axes if num_axes == -1 else start_axis + num_axes
+    assert end_axis <= input_num_axes, "end_axis[%d] = axis[%d] + num_axes[%d] is out of range"\
+            % (end_axis, start_axis, num_axes)
+
+    num_axes_replaced = end_axis - start_axis
+    num_axes_retained = input_num_axes - num_axes_replaced
+    num_new_axes = len(list(params.shape.dim))
+    output_shape = []
+
+    for i in range(start_axis):
+        output_shape.append(inshape[i])
+
+    for i in range(num_new_axes):
+        output_shape.append(params.shape.dim[i])
+
+    for i in range(end_axis, input_num_axes):
+        output_shape.append(inshape[i])
+
+    assert len(output_shape) == num_axes_retained + num_new_axes,\
+            "[Reshape]invalid dims of output shape[%s]" % (str(output_shape))
+
+    inferred_axis = -1
+    copy_axes = []
+    constant_count = 1
+    for i in range(num_new_axes):
+        top_dim = params.shape.dim[i]
+        if top_dim == 0:
+            copy_axes.append(i)
+            copy_axis_index = start_axis + i
+            output_shape[copy_axis_index] = inshape[copy_axis_index]
+        elif top_dim == -1:
+            assert inferred_axis == -1, "[Reshape]new shape contains multiple -1 dims"
+            inferred_axis = i
+        else:
+            constant_count *= top_dim
+
+    if inferred_axis >= 0:
+        explicit_count = constant_count
+        l = inshape[0:start_axis]
+        if len(l) > 0:
+            explicit_count *= count(l)
+        l = inshape[end_axis:]
+        if len(l) > 0:
+            explicit_count *= count(l)
+        for i in range(len(copy_axes)):
+            explicit_count *= output_shape[start_axis + copy_axes[i]]
+        assert input_count % explicit_count == 0, "[Reshape]botom count[%d] "\
+                "must be divisible by product of the specified dimensions[%d] "\
+                % (input_count, explicit_count)
+        output_shape[start_axis + inferred_axis] = int(input_count / explicit_count)
+
+    output_count = count(output_shape)
+    assert output_count == input_count, "[Reshape]output count[%d] must match input count[%d]" % (
+        output_count, input_count)
+    output_shape[0] = -1
+    return [output_shape]
+
+
+def shape_argmax(layer, input_shape):
+    inshape = input_shape[0]
+    params = layer.argmax_param
+    out_max_val = params.out_max_val if hasattr(params, out_max_val) else False
+    top_k = params.top_k if hasattr(params, top_k) else 1
+    axis = parmas.axis if hasattr(params, axis) else -1
+    if axis < 0:
+        axis += len(inshape)
+    assert (axis + 1 == len(inshape)
+            ), 'only can be applied on the last dimension[axis:%d, %s] now,'\
+                    'make sure you have set axis param in xxx.prototxt file' \
+                    % (axis, str(inshape))
+
+    output_shape = inshape
+    output_shape[-1] = top_k
+    if out_max_val is True:
+        output_shape[-1] *= 2
+    return [output_shape]
+
+
+def shape_crop(layer, input_shape):
+    assert len(input_shape) == 2, "the number of crop's inputs must be 2"
+    return [input_shape[1]]
+
+
+def shape_flatten(layer, input_shape):
+    assert len(input_shape) == 1, "the number of flatten's inputs must be 1"
+    inshape = input_shape[0]
+    params = layer.flatten_param
+    start_axis = params.axis
+    end_axis = params.end_axis
+    if start_axis < 0:
+        start_axis += len(inshape)
+    if end_axis < 0:
+        end_axis += len(inshape) + 1
+    assert start_axis <= end_axis, 'invalid axis[%d] or end_axis[%d] params'\
+            % (start_axis, end_axis)
+    output_shape = inshape[0:start_axis]
+    if len(inshape[start_axis:end_axis]) != 0:
+        flat_sz = reduce(lambda a, b: a * b, inshape[start_axis:end_axis])
+        output_shape += [flat_sz]
+    output_shape += inshape[end_axis:len(inshape)]
+    output_shape[0] = -1
+    return [output_shape]
+
+
+def shape_power(layer, input_shape):
+    return input_shape
+
+
+def shape_reduction(layer, input_shape):
+    params = layer.reduction_param
+    axis = params.axis
+    if axis < 0:
+        axis += len(input_shape[0]) + 1
+    assert axis <= len(input_shape[0]), 'invalid axis[%d] error' % (axis)
+    return [input_shape[0:axis]]
+
+def shape_axpy(layer, input_shape):
+    assert len(input_shapes) == 3, "not valid input shape for axpy layer"
+    assert len(input_shapes[0]) == len(input_shapes[1]), 'should have same dims'
+    output_shape = input_shapes[1]
+    assert (input_shapes[2] == output_shape),\
+            "shape not consistent for axpy[%s <--> %s]" \
+            % (str(output_shape), str(input_shapes[2]))
+    return [output_shape]
+
+def shape_detectionoutput(layer, input_shape):
+    return [[-1, 6]]
+
+def shape_normalize(layer, input_shape):
+    return input_shape
+
+def shape_permute(layer, input_shape, order=None):
+    inshape = input_shape[0]
+    output_shape = []
+    order = list(order)
+    for ii in order:
+        assert ii < len(inshape), "invalid order for permute[%s]" % (name)
+        output_shape.append(inshape[ii])
+    return [output_shape]
+
+def shape_priorbox(layer, input_shape, max_size=None, aspect_ratio=None):
+    fc_shape = input_shape[0]
+    N = 1
+    if not max_size == None:
+        N += 1
+    if not aspect_ratio == None:
+        N += 2 * len(aspect_ratio)
+    N_bbx = fc_shape[2] * fc_shape[3] * N
+    output_shape = [1, 2, 4 * N_bbx]
+    return [output_shape]
+
+def shape_relu6(layer, input_shape):
+    return input_shape
+
+def shape_roipooling(layer, input_shape, pooled_w=None, pooled_h=None):
+    base_fea_shape = input_shapes[0]
+    rois_shape = input_shapes[1]
+    output_shape = base_fea_shape
+    output_shape[0] = rois_shape[0]
+    output_shape[2] = pooled_h
+    output_shape[3] = pooled_w
+    return [output_shape]
+
+def shape_shufflechannel(layer, input_shape):
+    return input_shape
+
+def shape_upsample(layer, input_shape, scale):
+    assert len(input_shapes) == 1, "not valid input shape for upsample layer"
+    assert type(scale) is int
+    input_shape = input_shapes[0]
+    new_h = scale * input_shape[2]
+    new_w = scale * input_shape[3]
+
+    output_shape = [input_shape[0], input_shape[1], new_h, new_w]
+    return [output_shape]
+
+def shape_select(layer, input_shape, slice_point, axis):
+    input_shape = input_shapes[0]
+    start = slice_point[0]
+    if len(slice_point) == 2:
+        end = slice_point[1]
+    else:
+        end = input_shape[axis]
+    assert end > start, "invalid slice_point with [start:%d, end:%d]"\
+             % (start, end)
+    output_shape = input_shape
+    output_shape[axis] = end - start
+    return [output_shape]
+

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/axpy.py

@@ -16,8 +16,8 @@ def axpy_layer(inputs, input_shape=None, name=None):
     alpha = inputs[0]
     x = inputs[1]
     y = inputs[2]
-    out = fluid.layers.elementwise_mul(x, alpha, axis=0)
-    out = fluid.layers.elementwise_add(out, y, name=name)
+    out = paddle.multiply(x, alpha, axis=0)
+    out = paddle.add(out, y, name=name)
     return out
 
 

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/convolutiondepthwise.py

@@ -15,15 +15,6 @@ def convolutiondepthwise_shape(input_shape,
                                kernel_w=None,
                                stride_h=None,
                                stride_w=None):
-    [k_h, k_w] = [1, 1]
-    if isinstance(kernel_size, numbers.Number):
-        [k_h, k_w] = [kernel_size] * 2
-    elif len(kernel_size) > 0:
-        k_h = kernel_h if kernel_h > 0 else kernel_size[0]
-        k_w = kernel_w if kernel_w > 0 else kernel_size[len(kernel_size) - 1]
-    elif kernel_h > 0 or kernel_w > 0:
-        k_h = kernel_h
-        k_w = kernel_w
     [s_h, s_w] = [1, 1]
     if isinstance(stride, numbers.Number):
         [s_h, s_w] = [stride] * 2
@@ -79,15 +70,6 @@ def convolutiondepthwise_layer(inputs,
                                input_shape=None,
                                name=None):
     import numbers
-    [k_h, k_w] = [1, 1]
-    if isinstance(kernel_size, numbers.Number):
-        [k_h, k_w] = [kernel_size] * 2
-    elif len(kernel_size) > 0:
-        k_h = kernel_h if kernel_h > 0 else kernel_size[0]
-        k_w = kernel_w if kernel_w > 0 else kernel_size[len(kernel_size) - 1]
-    elif kernel_h > 0 or kernel_w > 0:
-        k_h = kernel_h
-        k_w = kernel_w
     [s_h, s_w] = [1, 1]
     if isinstance(stride, numbers.Number):
         [s_h, s_w] = [stride] * 2
@@ -122,16 +104,14 @@ def convolutiondepthwise_layer(inputs,
     c_out = num_output if num_output is not None else input_shape[0][1]
     group = int(c_in / (c_in / c_out)) if c_in > c_out else int(c_in /
                                                                 (c_out / c_in))
-    out = fluid.layers.conv2d(
+    out = paddle.nn.functional.conv2d(
         input,
         dilation=[dila_h, dila_w],
-        filter_size=[k_h, k_w],
         stride=[s_h, s_w],
         padding=[p_h, p_w],
         groups=group,
-        num_filters=c_out,
-        param_attr=name + '_weights',
-        bias_attr=name + '_bias',
+        weight=name + '_weights',
+        bias=name + '_bias',
         name=name)
     return out
 

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/detectionoutput.py

@@ -14,30 +14,18 @@ def detectionoutput_layer(inputs,
                           confidence_threshold=0.1,
                           input_shape=None,
                           name=None):
-    nms_param_str = nms_param
-    nms_param = {}
-    part = nms_param_str.split(',')
-    for s in part:
-        if s == '':
-            break
-        else:
-            name, obj = s.split(': ')
-            if name == 'top_k':
-                nms_param[name] = int(obj)
-            else:
-                nms_param[name] = float(obj)
     if nms_param is None:
         nms_param = {"nms_threshold": 0.3, "top_k": 10, "eta": 1.0}
     mbox_conf_flatten = inputs[1]
     mbox_priorbox = inputs[2]
-    mbox_priorbox_list = fluid.layers.split(mbox_priorbox, 2, dim=1)
+    mbox_priorbox_list = paddle.split(mbox_priorbox, 2, dim=1)
     pb = mbox_priorbox_list[0]
     pbv = mbox_priorbox_list[1]
-    pb = fluid.layers.reshape(x=pb, shape=[-1, 4])
-    pbv = fluid.layers.reshape(x=pbv, shape=[-1, 4])
+    pb = paddle.reshape(x=pb, shape=[-1, 4])
+    pbv = paddle.reshape(x=pbv, shape=[-1, 4])
     mbox_loc = inputs[0]
-    mbox_loc = fluid.layers.reshape(x=mbox_loc, shape=[-1, pb.shape[0], 4])
-    mbox_conf_flatten = fluid.layers.reshape(
+    mbox_loc = paddle.reshape(x=mbox_loc, shape=[-1, pb.shape[0], 4])
+    mbox_conf_flatten = paddle.reshape(
         x=mbox_conf_flatten, shape=[0, pb.shape[0], -1])
 
     default = {"nms_threshold": 0.3, "top_k": 10, "eta": 1.0}

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/normalize.py

@@ -13,14 +13,14 @@ def normalize_layer(inputs,
                     name=None):
     assert across_spatial == False, "Only support across_spatial == False for Normalize"
     input = inputs[0]
-    l2_norm = fluid.layers.l2_normalize(input, axis=1, name=name + '_l2')
-    scale_param = fluid.layers.create_parameter(
+    l2_norm = paddle.nn.functional.normalize(input, axis=1, p=2, name=name + '_l2')
+    scale_param = paddle.static.create_parameter(
         shape=[1] if channel_shared else [1, 1, 1, input_shape[0][1]],
         dtype=input.dtype,
-        attr=fluid.ParamAttr(name=name + '_scale'))
-    scale_param = fluid.layers.reshape(x=scale_param, \
+        attr=paddle.ParamAttr(name=name + '_scale'))
+    scale_param = paddle.reshape(x=scale_param, \
                   shape=[1] if channel_shared else [input_shape[0][1]])
-    out = fluid.layers.elementwise_mul(
+    out = paddle.multiply(
         x=l2_norm, y=scale_param, axis=-1 if channel_shared else 1)
     return out
 

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/permute.py

@@ -14,7 +14,7 @@ def permute_shape(input_shape, order=None):
 def permute_layer(inputs, order=None, input_shape=None, name=None):
     input = inputs[0]
     order = list(order)
-    out = fluid.layers.transpose(input, perm=order, name=name)
+    out = paddle.transpose(input, perm=order, name=name)
     return out
 
 

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/relu6.py

@@ -8,7 +8,7 @@ def relu6_shape(input_shape):
 
 def relu6_layer(inputs, input_shape=None, name=None):
     input = inputs[0]
-    out = fluid.layers.relu6(x=input)
+    out = paddle.nn.functional.relu6(x=input)
     return out
 
 

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/roipooling.py

@@ -20,7 +20,7 @@ def roipooling_layer(inputs,
                      name=None):
     input = inputs[0]
     roi = inputs[1]
-    roi = fluid.layers.slice(roi, axes=[1], starts=[1], ends=[5])
+    roi = paddle.slice(roi, axes=[1], starts=[1], ends=[5])
     out = fluid.layers.roi_pool(
         input,
         roi,

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/select.py

@@ -30,7 +30,7 @@ def select_layer(inputs,
     out = []
     for i in range(len(slice_point)):
         out.append(
-            fluid.layers.slice(
+            paddle.slice(
                 input,
                 axes=[axis],
                 starts=[slice_point[i]],

x2paddle/op_mapper/static/caffe2paddle/caffe_custom_layer/upsample.py

@@ -42,8 +42,8 @@ def upsample_layer(inputs, scale, input_shape=None, name=None):
     :return:
     """
     x = inputs[0]
-    out = fluid.layers.resize_nearest(
-        x, align_corners=False, scale=scale, name=name)
+    out = paddle.nn.functioanl.interpolate(
+        x, align_corners=False, scale_factor=scale, name=name)
 
     return out