Merge pull request #89 from qingqing01/cyclegan_ocr_update

Update OCR model

examples/ocr/eval.py

@@ -16,11 +16,11 @@ from __future__ import print_function
 import argparse
 import functools
 
-import paddle.fluid.profiler as profiler
+import paddle
 import paddle.fluid as fluid
 
-from paddle.incubate.hapi.model import Input, set_device
-from paddle.incubate.hapi.vision.transforms import BatchCompose
+from paddle.static import InputSpec as Input
+from paddle.vision.transforms import BatchCompose
 
 from utility import add_arguments, print_arguments
 from utility import SeqAccuracy, LoggerCallBack, SeqBeamAccuracy
@@ -46,13 +46,8 @@ add_arg('dynamic',           bool,  False,              "Whether to use dygraph.
 
 
 def main(FLAGS):
-    device = set_device("gpu" if FLAGS.use_gpu else "cpu")
+    device = paddle.set_device("gpu" if FLAGS.use_gpu else "cpu")
     fluid.enable_dygraph(device) if FLAGS.dynamic else None
-    model = Seq2SeqAttModel(
-        encoder_size=FLAGS.encoder_size,
-        decoder_size=FLAGS.decoder_size,
-        emb_dim=FLAGS.embedding_dim,
-        num_classes=FLAGS.num_classes)
 
     # yapf: disable
     inputs = [
@@ -64,13 +59,16 @@ def main(FLAGS):
         Input([None, None], "float32", name="mask")
     ]
     # yapf: enable
-
-    model.prepare(
-        loss_function=WeightCrossEntropy(),
-        metrics=SeqAccuracy(),
+    model = paddle.Model(
+        Seq2SeqAttModel(
+            encoder_size=FLAGS.encoder_size,
+            decoder_size=FLAGS.decoder_size,
+            emb_dim=FLAGS.embedding_dim,
+            num_classes=FLAGS.num_classes),
         inputs=inputs,
-        labels=labels,
-        device=device)
+        labels=labels)
+
+    model.prepare(loss=WeightCrossEntropy(), metrics=SeqAccuracy())
     model.load(FLAGS.init_model)
 
     test_dataset = data.test()
@@ -97,29 +95,29 @@ def main(FLAGS):
 def beam_search(FLAGS):
     device = set_device("gpu" if FLAGS.use_gpu else "cpu")
     fluid.enable_dygraph(device) if FLAGS.dynamic else None
-    model = Seq2SeqAttInferModel(
-        encoder_size=FLAGS.encoder_size,
-        decoder_size=FLAGS.decoder_size,
-        emb_dim=FLAGS.embedding_dim,
-        num_classes=FLAGS.num_classes,
-        beam_size=FLAGS.beam_size)
 
+    # yapf: disable
     inputs = [
-        Input(
-            [None, 1, 48, 384], "float32", name="pixel"), Input(
-                [None, None], "int64", name="label_in")
+        Input([None, 1, 48, 384], "float32", name="pixel"),
+        Input([None, None], "int64", name="label_in")
     ]
     labels = [
-        Input(
-            [None, None], "int64", name="label_out"), Input(
-                [None, None], "float32", name="mask")
+        Input([None, None], "int64", name="label_out"),
+        Input([None, None], "float32", name="mask")
     ]
-    model.prepare(
-        loss_function=None,
-        metrics=SeqBeamAccuracy(),
+    # yapf: enable
+
+    model = paddle.Model(
+        Seq2SeqAttInferModel(
+            encoder_size=FLAGS.encoder_size,
+            decoder_size=FLAGS.decoder_size,
+            emb_dim=FLAGS.embedding_dim,
+            num_classes=FLAGS.num_classes,
+            beam_size=FLAGS.beam_size),
         inputs=inputs,
-        labels=labels,
-        device=device)
+        labels=labels)
+
+    model.prepare(loss_function=None, metrics=SeqBeamAccuracy())
     model.load(FLAGS.init_model)
 
     test_dataset = data.test()

examples/ocr/predict.py

@@ -22,12 +22,12 @@ import argparse
 import functools
 from PIL import Image
 
-import paddle.fluid.profiler as profiler
+import paddle
 import paddle.fluid as fluid
 
-from paddle.incubate.hapi.model import Input, set_device
-from paddle.incubate.hapi.datasets.folder import ImageFolder
-from paddle.incubate.hapi.vision.transforms import BatchCompose
+from paddle.static import InputSpec as Input
+from paddle.vision.datasets.folder import ImageFolder
+from paddle.vision.transforms import BatchCompose
 
 from utility import add_arguments, print_arguments
 from utility import postprocess, index2word
@@ -52,18 +52,20 @@ add_arg('dynamic',           bool,  False,   "Whether to use dygraph.")
 
 
 def main(FLAGS):
-    device = set_device("gpu" if FLAGS.use_gpu else "cpu")
+    device = paddle.set_device("gpu" if FLAGS.use_gpu else "cpu")
     fluid.enable_dygraph(device) if FLAGS.dynamic else None
-    model = Seq2SeqAttInferModel(
-        encoder_size=FLAGS.encoder_size,
-        decoder_size=FLAGS.decoder_size,
-        emb_dim=FLAGS.embedding_dim,
-        num_classes=FLAGS.num_classes,
-        beam_size=FLAGS.beam_size)
 
     inputs = [Input([None, 1, 48, 384], "float32", name="pixel"), ]
+    model = paddle.Model(
+        Seq2SeqAttInferModel(
+            encoder_size=FLAGS.encoder_size,
+            decoder_size=FLAGS.decoder_size,
+            emb_dim=FLAGS.embedding_dim,
+            num_classes=FLAGS.num_classes,
+            beam_size=FLAGS.beam_size),
+        inputs)
 
-    model.prepare(inputs=inputs, device=device)
+    model.prepare()
     model.load(FLAGS.init_model)
 
     fn = lambda p: Image.open(p).convert('L')

examples/ocr/seq2seq_attn.py

@@ -15,16 +15,15 @@ from __future__ import print_function
 
 import numpy as np
 
+import paddle
 import paddle.fluid as fluid
 import paddle.fluid.layers as layers
 from paddle.fluid.layers import BeamSearchDecoder
 
-from paddle.incubate.hapi.text import RNNCell, RNN, DynamicDecode
-from paddle.incubate.hapi.model import Model
-from paddle.incubate.hapi.loss import Loss
+from paddle.text import RNNCell, RNN, DynamicDecode
 
 
-class ConvBNPool(fluid.dygraph.Layer):
+class ConvBNPool(paddle.nn.Layer):
     def __init__(self,
                  in_ch,
                  out_ch,
@@ -83,7 +82,7 @@ class ConvBNPool(fluid.dygraph.Layer):
         return out
 
 
-class CNN(fluid.dygraph.Layer):
+class CNN(paddle.nn.Layer):
     def __init__(self, in_ch=1, is_test=False):
         super(CNN, self).__init__()
         self.conv_bn1 = ConvBNPool(in_ch, 16)
@@ -136,7 +135,7 @@ class GRUCell(RNNCell):
         return [self.hidden_size]
 
 
-class Encoder(fluid.dygraph.Layer):
+class Encoder(paddle.nn.Layer):
     def __init__(
             self,
             in_channel=1,
@@ -187,7 +186,7 @@ class Encoder(fluid.dygraph.Layer):
         return gru_bwd, encoded_vector, encoded_proj
 
 
-class Attention(fluid.dygraph.Layer):
+class Attention(paddle.nn.Layer):
     """
     Neural Machine Translation by Jointly Learning to Align and Translate.
     https://arxiv.org/abs/1409.0473
@@ -232,7 +231,7 @@ class DecoderCell(RNNCell):
         return hidden, hidden
 
 
-class Decoder(fluid.dygraph.Layer):
+class Decoder(paddle.nn.Layer):
     def __init__(self, num_classes, emb_dim, encoder_size, decoder_size):
         super(Decoder, self).__init__()
         self.decoder_attention = RNN(DecoderCell(encoder_size, decoder_size))
@@ -249,7 +248,7 @@ class Decoder(fluid.dygraph.Layer):
         return pred
 
 
-class Seq2SeqAttModel(Model):
+class Seq2SeqAttModel(paddle.nn.Layer):
     def __init__(
             self,
             in_channle=1,
@@ -322,12 +321,11 @@ class Seq2SeqAttInferModel(Seq2SeqAttModel):
         return rs
 
 
-class WeightCrossEntropy(Loss):
+class WeightCrossEntropy(paddle.nn.Layer):
     def __init__(self):
-        super(WeightCrossEntropy, self).__init__(average=False)
+        super(WeightCrossEntropy, self).__init__()
 
-    def forward(self, outputs, labels):
-        predict, (label, mask) = outputs[0], labels
+    def forward(self, predict, label, mask):
         loss = layers.cross_entropy(predict, label=label)
         loss = layers.elementwise_mul(loss, mask, axis=0)
         loss = layers.reduce_sum(loss)

examples/ocr/train.py

@@ -21,11 +21,11 @@ import numpy as np
 import argparse
 import functools
 
-import paddle.fluid.profiler as profiler
+import paddle
 import paddle.fluid as fluid
 
-from paddle.incubate.hapi.model import Input, set_device
-from paddle.incubate.hapi.vision.transforms import BatchCompose
+from paddle.static import InputSpec as Input
+from paddle.vision.transforms import BatchCompose
 
 from utility import add_arguments, print_arguments
 from utility import SeqAccuracy, LoggerCallBack
@@ -58,14 +58,28 @@ add_arg('dynamic',           bool,  False,      "Whether to use dygraph.")
 
 
 def main(FLAGS):
-    device = set_device("gpu" if FLAGS.use_gpu else "cpu")
+    device = paddle.set_device("gpu" if FLAGS.use_gpu else "cpu")
     fluid.enable_dygraph(device) if FLAGS.dynamic else None
 
-    model = Seq2SeqAttModel(
-        encoder_size=FLAGS.encoder_size,
-        decoder_size=FLAGS.decoder_size,
-        emb_dim=FLAGS.embedding_dim,
-        num_classes=FLAGS.num_classes)
+    # yapf: disable
+    inputs = [
+        Input([None,1,48,384], "float32", name="pixel"),
+        Input([None, None], "int64", name="label_in"),
+    ]
+    labels = [
+        Input([None, None], "int64", name="label_out"),
+        Input([None, None], "float32", name="mask"),
+    ]
+    # yapf: enable
+
+    model = paddle.Model(
+        Seq2SeqAttModel(
+            encoder_size=FLAGS.encoder_size,
+            decoder_size=FLAGS.decoder_size,
+            emb_dim=FLAGS.embedding_dim,
+            num_classes=FLAGS.num_classes),
+        inputs,
+        labels)
 
     lr = FLAGS.lr
     if FLAGS.lr_decay_strategy == "piecewise_decay":
@@ -79,23 +93,7 @@ def main(FLAGS):
         parameter_list=model.parameters(),
         grad_clip=grad_clip)
 
-    # yapf: disable
-    inputs = [
-        Input([None,1,48,384], "float32", name="pixel"),
-        Input([None, None], "int64", name="label_in"),
-    ]
-    labels = [
-        Input([None, None], "int64", name="label_out"),
-        Input([None, None], "float32", name="mask"),
-    ]
-    # yapf: enable
-
-    model.prepare(
-        optimizer,
-        WeightCrossEntropy(),
-        SeqAccuracy(),
-        inputs=inputs,
-        labels=labels)
+    model.prepare(optimizer, WeightCrossEntropy(), SeqAccuracy())
 
     train_dataset = data.train()
     train_collate_fn = BatchCompose(

examples/ocr/utility.py

@@ -18,11 +18,12 @@ from __future__ import division
 from __future__ import print_function
 import distutils.util
 import numpy as np
-import paddle.fluid as fluid
 import six
 
-from paddle.incubate.hapi.metrics import Metric
-from paddle.incubate.hapi.callbacks import ProgBarLogger
+import paddle
+import paddle.fluid as fluid
+
+from paddle.metric import Metric
 
 
 def print_arguments(args):
@@ -72,7 +73,7 @@ class SeqAccuracy(Metric):
         self._name = 'seq_acc'
         self.reset()
 
-    def add_metric_op(self, output, label, mask, *args, **kwargs):
+    def compute(self, output, label, mask, *args, **kwargs):
         pred = fluid.layers.flatten(output, axis=2)
         score, topk = fluid.layers.topk(pred, 1)
         return topk, label, mask
@@ -102,7 +103,7 @@ class SeqAccuracy(Metric):
         return self._name
 
 
-class LoggerCallBack(ProgBarLogger):
+class LoggerCallBack(paddle.callbacks.ProgBarLogger):
     def __init__(self, log_freq=1, verbose=2, train_bs=None, eval_bs=None):
         super(LoggerCallBack, self).__init__(log_freq, verbose)
         self.train_bs = train_bs
@@ -153,7 +154,7 @@ class SeqBeamAccuracy(Metric):
         self._name = 'seq_acc'
         self.reset()
 
-    def add_metric_op(self, output, label, mask, *args, **kwargs):
+    def compute(self, output, label, mask, *args, **kwargs):
         return output, label, mask
 
     def update(self, preds, labels, masks, *args, **kwargs):