merge dygraph

ppocr/losses/combined_loss.py

@@ -49,11 +49,15 @@ class CombinedLoss(nn.Layer):
             loss = loss_func(input, batch, **kargs)
             if isinstance(loss, paddle.Tensor):
                 loss = {"loss_{}_{}".format(str(loss), idx): loss}
+
             weight = self.loss_weight[idx]
-            for key in loss.keys():
-                if key == "loss":
-                    loss_all += loss[key] * weight
-                else:
-                    loss_dict["{}_{}".format(key, idx)] = loss[key]
+
+            loss = {key: loss[key] * weight for key in loss}
+
+            if "loss" in loss:
+                loss_all += loss["loss"]
+            else:
+                loss_all += paddle.add_n(list(loss.values()))
+            loss_dict.update(loss)
         loss_dict["loss"] = loss_all
         return loss_dict

ppocr/losses/distillation_loss.py

@@ -44,20 +44,22 @@ class DistillationDMLLoss(DMLLoss):
     def __init__(self,
                  model_name_pairs=[],
                  act=None,
+                 use_log=False,
                  key=None,
                  maps_name=None,
                  name="dml"):
-        super().__init__(act=act)
+        super().__init__(act=act, use_log=use_log)
         assert isinstance(model_name_pairs, list)
         self.key = key
         self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
         self.name = name
         self.maps_name = self._check_maps_name(maps_name)
-    
+
     def _check_model_name_pairs(self, model_name_pairs):
         if not isinstance(model_name_pairs, list):
             return []
-        elif isinstance(model_name_pairs[0], list) and isinstance(model_name_pairs[0][0], str):
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+                model_name_pairs[0][0], str):
             return model_name_pairs
         else:
             return [model_name_pairs]
@@ -112,9 +114,9 @@ class DistillationDMLLoss(DMLLoss):
                             loss_dict["{}_{}_{}_{}_{}".format(key, pair[
                                 0], pair[1], map_name, idx)] = loss[key]
                     else:
-                        loss_dict["{}_{}_{}".format(self.name, self.maps_name[_c],
-                                                    idx)] = loss
-        
+                        loss_dict["{}_{}_{}".format(self.name, self.maps_name[
+                            _c], idx)] = loss
+
         loss_dict = _sum_loss(loss_dict)
 
         return loss_dict

ppocr/losses/rec_sar_loss.py

+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class SARLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(SARLoss, self).__init__()
+        self.loss_func = paddle.nn.loss.CrossEntropyLoss(reduction="mean", ignore_index=96)
+
+    def forward(self, predicts, batch):
+        predict = predicts[:, :-1, :] # ignore last index of outputs to be in same seq_len with targets
+        label = batch[1].astype("int64")[:, 1:] # ignore first index of target in loss calculation
+        batch_size, num_steps, num_classes = predict.shape[0], predict.shape[
+            1], predict.shape[2]
+        assert len(label.shape) == len(list(predict.shape)) - 1, \
+            "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
+
+        inputs = paddle.reshape(predict, [-1, num_classes])
+        targets = paddle.reshape(label, [-1])
+        loss = self.loss_func(inputs, targets)
+        return {'loss': loss}

ppocr/modeling/backbones/__init__.py

@@ -27,8 +27,9 @@ def build_backbone(config, model_type):
         from .rec_resnet_fpn import ResNetFPN
         from .rec_mv1_enhance import MobileNetV1Enhance
         from .rec_nrtr_mtb import MTB
+        from .rec_resnet_31 import ResNet31
         support_dict = [
-            'MobileNetV1Enhance', 'MobileNetV3', 'ResNet', 'ResNetFPN', 'MTB'
+            'MobileNetV1Enhance', 'MobileNetV3', 'ResNet', 'ResNetFPN', 'MTB', "ResNet31"
         ]
     elif model_type == "e2e":
         from .e2e_resnet_vd_pg import ResNet

ppocr/modeling/backbones/rec_resnet_31.py

+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+
+__all__ = ["ResNet31"]
+
+
+def conv3x3(in_channel, out_channel, stride=1):
+    return nn.Conv2D(
+        in_channel,
+        out_channel,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        bias_attr=False
+    )
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+    def __init__(self, in_channels, channels, stride=1, downsample=False):
+        super().__init__()
+        self.conv1 = conv3x3(in_channels, channels, stride)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(channels, channels)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.downsample = downsample
+        if downsample:
+            self.downsample = nn.Sequential(
+                nn.Conv2D(in_channels, channels * self.expansion, 1, stride, bias_attr=False),
+                nn.BatchNorm2D(channels * self.expansion),
+            )
+        else:
+            self.downsample = nn.Sequential()
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out        
+
+
+class ResNet31(nn.Layer):
+    '''
+    Args:
+        in_channels (int): Number of channels of input image tensor.
+        layers (list[int]): List of BasicBlock number for each stage.
+        channels (list[int]): List of out_channels of Conv2d layer.
+        out_indices (None | Sequence[int]): Indices of output stages.
+        last_stage_pool (bool): If True, add `MaxPool2d` layer to last stage.
+    '''
+    def __init__(self, 
+                in_channels=3, 
+                layers=[1, 2, 5, 3],
+                channels=[64, 128, 256, 256, 512, 512, 512],
+                out_indices=None,
+                last_stage_pool=False):
+        super(ResNet31, self).__init__()
+        assert isinstance(in_channels, int)
+        assert isinstance(last_stage_pool, bool)
+
+        self.out_indices = out_indices
+        self.last_stage_pool = last_stage_pool
+
+        # conv 1 (Conv Conv)
+        self.conv1_1 = nn.Conv2D(in_channels, channels[0], kernel_size=3, stride=1, padding=1)
+        self.bn1_1 = nn.BatchNorm2D(channels[0])
+        self.relu1_1 = nn.ReLU()
+
+        self.conv1_2 = nn.Conv2D(channels[0], channels[1], kernel_size=3, stride=1, padding=1)
+        self.bn1_2 = nn.BatchNorm2D(channels[1])
+        self.relu1_2 = nn.ReLU()
+
+        # conv 2 (Max-pooling, Residual block, Conv)
+        self.pool2 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block2 = self._make_layer(channels[1], channels[2], layers[0])
+        self.conv2 = nn.Conv2D(channels[2], channels[2], kernel_size=3, stride=1, padding=1)
+        self.bn2 = nn.BatchNorm2D(channels[2])
+        self.relu2 = nn.ReLU()
+
+        # conv 3 (Max-pooling, Residual block, Conv)
+        self.pool3 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block3 = self._make_layer(channels[2], channels[3], layers[1])
+        self.conv3 = nn.Conv2D(channels[3], channels[3], kernel_size=3, stride=1, padding=1)
+        self.bn3 = nn.BatchNorm2D(channels[3])
+        self.relu3 = nn.ReLU()
+
+        # conv 4 (Max-pooling, Residual block, Conv)
+        self.pool4 = nn.MaxPool2D(kernel_size=(2, 1), stride=(2, 1), padding=0, ceil_mode=True)
+        self.block4 = self._make_layer(channels[3], channels[4], layers[2])
+        self.conv4 = nn.Conv2D(channels[4], channels[4], kernel_size=3, stride=1, padding=1)
+        self.bn4 = nn.BatchNorm2D(channels[4])
+        self.relu4 = nn.ReLU()
+
+        # conv 5 ((Max-pooling), Residual block, Conv)
+        self.pool5 = None
+        if self.last_stage_pool:
+            self.pool5 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block5 = self._make_layer(channels[4], channels[5], layers[3])
+        self.conv5 = nn.Conv2D(channels[5], channels[5], kernel_size=3, stride=1, padding=1)
+        self.bn5 = nn.BatchNorm2D(channels[5])
+        self.relu5 = nn.ReLU()
+
+        self.out_channels = channels[-1]
+    
+    def _make_layer(self, input_channels, output_channels, blocks):
+        layers = []
+        for _ in range(blocks):
+            downsample = None
+            if input_channels != output_channels:
+                downsample = nn.Sequential(
+                    nn.Conv2D(
+                        input_channels, 
+                        output_channels, 
+                        kernel_size=1, 
+                        stride=1, 
+                        bias_attr=False),
+                    nn.BatchNorm2D(output_channels),
+                )
+                
+            layers.append(BasicBlock(input_channels, output_channels, downsample=downsample))
+            input_channels = output_channels
+        return nn.Sequential(*layers)
+
+
+    def forward(self, x):
+        x = self.conv1_1(x)
+        x = self.bn1_1(x)
+        x = self.relu1_1(x)
+
+        x = self.conv1_2(x)
+        x = self.bn1_2(x)
+        x = self.relu1_2(x)
+
+        outs = []
+        for i in range(4):
+            layer_index = i + 2
+            pool_layer = getattr(self, f'pool{layer_index}')
+            block_layer = getattr(self, f'block{layer_index}')
+            conv_layer = getattr(self, f'conv{layer_index}')
+            bn_layer = getattr(self, f'bn{layer_index}')
+            relu_layer = getattr(self, f'relu{layer_index}')
+
+            if pool_layer is not None:
+                x = pool_layer(x)
+            x = block_layer(x)
+            x = conv_layer(x)
+            x = bn_layer(x)
+            x= relu_layer(x)
+
+            outs.append(x)
+        
+        if self.out_indices is not None:
+            return tuple([outs[i] for i in self.out_indices])
+        
+        return x

ppocr/modeling/heads/__init__.py

@@ -28,13 +28,14 @@ def build_head(config):
     from .rec_att_head import AttentionHead
     from .rec_srn_head import SRNHead
     from .rec_nrtr_head import Transformer
+    from .rec_sar_head import SARHead
 
     # cls head
     from .cls_head import ClsHead
     support_dict = [
         'DBHead', 'PSEHead', 'EASTHead', 'SASTHead', 'CTCHead', 'ClsHead',
         'AttentionHead', 'SRNHead', 'PGHead', 'Transformer',
-        'TableAttentionHead'
+        'TableAttentionHead', 'SARHead'
     ]
 
     #table head

ppocr/modeling/heads/rec_sar_head.py

+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class SAREncoder(nn.Layer):
+    """
+    Args:
+        enc_bi_rnn (bool): If True, use bidirectional RNN in encoder.
+        enc_drop_rnn (float): Dropout probability of RNN layer in encoder.
+        enc_gru (bool): If True, use GRU, else LSTM in encoder.
+        d_model (int): Dim of channels from backbone.
+        d_enc (int): Dim of encoder RNN layer.
+        mask (bool): If True, mask padding in RNN sequence.
+    """
+
+    def __init__(self,
+                 enc_bi_rnn=False,
+                 enc_drop_rnn=0.1,
+                 enc_gru=False,
+                 d_model=512,
+                 d_enc=512,
+                 mask=True,
+                 **kwargs):
+        super().__init__()
+        assert isinstance(enc_bi_rnn, bool)
+        assert isinstance(enc_drop_rnn, (int, float))
+        assert 0 <= enc_drop_rnn < 1.0
+        assert isinstance(enc_gru, bool)
+        assert isinstance(d_model, int)
+        assert isinstance(d_enc, int)
+        assert isinstance(mask, bool)
+
+        self.enc_bi_rnn = enc_bi_rnn
+        self.enc_drop_rnn = enc_drop_rnn
+        self.mask = mask
+
+        # LSTM Encoder
+        if enc_bi_rnn:
+            direction = 'bidirectional'
+        else:
+            direction = 'forward'
+        kwargs = dict(
+            input_size=d_model,
+            hidden_size=d_enc,
+            num_layers=2,
+            time_major=False,
+            dropout=enc_drop_rnn,
+            direction=direction)
+        if enc_gru:
+            self.rnn_encoder = nn.GRU(**kwargs)
+        else:
+            self.rnn_encoder = nn.LSTM(**kwargs)
+
+        # global feature transformation
+        encoder_rnn_out_size = d_enc * (int(enc_bi_rnn) + 1)
+        self.linear = nn.Linear(encoder_rnn_out_size, encoder_rnn_out_size)
+
+    def forward(self, feat, img_metas=None):
+        if img_metas is not None:
+            assert len(img_metas[0]) == feat.shape[0]
+
+        valid_ratios = None
+        if img_metas is not None and self.mask:
+            valid_ratios = img_metas[-1]
+
+        h_feat = feat.shape[2]  # bsz c h w
+        feat_v = F.max_pool2d(
+            feat, kernel_size=(h_feat, 1), stride=1, padding=0)
+        feat_v = feat_v.squeeze(2)  # bsz * C * W
+        feat_v = paddle.transpose(feat_v, perm=[0, 2, 1])  # bsz * W * C
+        holistic_feat = self.rnn_encoder(feat_v)[0]  # bsz * T * C
+
+        if valid_ratios is not None:
+            valid_hf = []
+            T = holistic_feat.shape[1]
+            for i, valid_ratio in enumerate(valid_ratios):
+                valid_step = min(T, math.ceil(T * valid_ratio)) - 1
+                valid_hf.append(holistic_feat[i, valid_step, :])
+            valid_hf = paddle.stack(valid_hf, axis=0)
+        else:
+            valid_hf = holistic_feat[:, -1, :]  # bsz * C
+        holistic_feat = self.linear(valid_hf)  # bsz * C
+
+        return holistic_feat
+
+
+class BaseDecoder(nn.Layer):
+    def __init__(self, **kwargs):
+        super().__init__()
+
+    def forward_train(self, feat, out_enc, targets, img_metas):
+        raise NotImplementedError
+
+    def forward_test(self, feat, out_enc, img_metas):
+        raise NotImplementedError
+
+    def forward(self,
+                feat,
+                out_enc,
+                label=None,
+                img_metas=None,
+                train_mode=True):
+        self.train_mode = train_mode
+
+        if train_mode:
+            return self.forward_train(feat, out_enc, label, img_metas)
+        return self.forward_test(feat, out_enc, img_metas)
+
+
+class ParallelSARDecoder(BaseDecoder):
+    """
+    Args:
+        out_channels (int): Output class number.
+        enc_bi_rnn (bool): If True, use bidirectional RNN in encoder.
+        dec_bi_rnn (bool): If True, use bidirectional RNN in decoder.
+        dec_drop_rnn (float): Dropout of RNN layer in decoder.
+        dec_gru (bool): If True, use GRU, else LSTM in decoder.
+        d_model (int): Dim of channels from backbone.
+        d_enc (int): Dim of encoder RNN layer.
+        d_k (int): Dim of channels of attention module.
+        pred_dropout (float): Dropout probability of prediction layer.
+        max_seq_len (int): Maximum sequence length for decoding.
+        mask (bool): If True, mask padding in feature map.
+        start_idx (int): Index of start token.
+        padding_idx (int): Index of padding token.
+        pred_concat (bool): If True, concat glimpse feature from
+            attention with holistic feature and hidden state.
+    """
+
+    def __init__(
+            self,
+            out_channels,  # 90 + unknown + start + padding
+            enc_bi_rnn=False,
+            dec_bi_rnn=False,
+            dec_drop_rnn=0.0,
+            dec_gru=False,
+            d_model=512,
+            d_enc=512,
+            d_k=64,
+            pred_dropout=0.1,
+            max_text_length=30,
+            mask=True,
+            pred_concat=True,
+            **kwargs):
+        super().__init__()
+
+        self.num_classes = out_channels
+        self.enc_bi_rnn = enc_bi_rnn
+        self.d_k = d_k
+        self.start_idx = out_channels - 2
+        self.padding_idx = out_channels - 1
+        self.max_seq_len = max_text_length
+        self.mask = mask
+        self.pred_concat = pred_concat
+
+        encoder_rnn_out_size = d_enc * (int(enc_bi_rnn) + 1)
+        decoder_rnn_out_size = encoder_rnn_out_size * (int(dec_bi_rnn) + 1)
+
+        # 2D attention layer
+        self.conv1x1_1 = nn.Linear(decoder_rnn_out_size, d_k)
+        self.conv3x3_1 = nn.Conv2D(
+            d_model, d_k, kernel_size=3, stride=1, padding=1)
+        self.conv1x1_2 = nn.Linear(d_k, 1)
+
+        # Decoder RNN layer
+        if dec_bi_rnn:
+            direction = 'bidirectional'
+        else:
+            direction = 'forward'
+
+        kwargs = dict(
+            input_size=encoder_rnn_out_size,
+            hidden_size=encoder_rnn_out_size,
+            num_layers=2,
+            time_major=False,
+            dropout=dec_drop_rnn,
+            direction=direction)
+        if dec_gru:
+            self.rnn_decoder = nn.GRU(**kwargs)
+        else:
+            self.rnn_decoder = nn.LSTM(**kwargs)
+
+        # Decoder input embedding
+        self.embedding = nn.Embedding(
+            self.num_classes,
+            encoder_rnn_out_size,
+            padding_idx=self.padding_idx)
+
+        # Prediction layer
+        self.pred_dropout = nn.Dropout(pred_dropout)
+        pred_num_classes = self.num_classes - 1
+        if pred_concat:
+            fc_in_channel = decoder_rnn_out_size + d_model + d_enc
+        else:
+            fc_in_channel = d_model
+        self.prediction = nn.Linear(fc_in_channel, pred_num_classes)
+
+    def _2d_attention(self,
+                      decoder_input,
+                      feat,
+                      holistic_feat,
+                      valid_ratios=None):
+
+        y = self.rnn_decoder(decoder_input)[0]
+        # y: bsz * (seq_len + 1) * hidden_size
+
+        attn_query = self.conv1x1_1(y)  # bsz * (seq_len + 1) * attn_size
+        bsz, seq_len, attn_size = attn_query.shape
+        attn_query = paddle.unsqueeze(attn_query, axis=[3, 4])
+        # (bsz, seq_len + 1, attn_size, 1, 1)
+
+        attn_key = self.conv3x3_1(feat)
+        # bsz * attn_size * h * w
+        attn_key = attn_key.unsqueeze(1)
+        # bsz * 1 * attn_size * h * w
+
+        attn_weight = paddle.tanh(paddle.add(attn_key, attn_query))
+
+        # bsz * (seq_len + 1) * attn_size * h * w
+        attn_weight = paddle.transpose(attn_weight, perm=[0, 1, 3, 4, 2])
+        # bsz * (seq_len + 1) * h * w * attn_size
+        attn_weight = self.conv1x1_2(attn_weight)
+        # bsz * (seq_len + 1) * h * w * 1
+        bsz, T, h, w, c = attn_weight.shape
+        assert c == 1
+
+        if valid_ratios is not None:
+            # cal mask of attention weight
+            for i, valid_ratio in enumerate(valid_ratios):
+                valid_width = min(w, math.ceil(w * valid_ratio))
+                attn_weight[i, :, :, valid_width:, :] = float('-inf')
+
+        attn_weight = paddle.reshape(attn_weight, [bsz, T, -1])
+        attn_weight = F.softmax(attn_weight, axis=-1)
+
+        attn_weight = paddle.reshape(attn_weight, [bsz, T, h, w, c])
+        attn_weight = paddle.transpose(attn_weight, perm=[0, 1, 4, 2, 3])
+        # attn_weight: bsz * T * c * h * w
+        # feat: bsz * c * h * w
+        attn_feat = paddle.sum(paddle.multiply(feat.unsqueeze(1), attn_weight),
+                               (3, 4),
+                               keepdim=False)
+        # bsz * (seq_len + 1) * C
+
+        # Linear transformation
+        if self.pred_concat:
+            hf_c = holistic_feat.shape[-1]
+            holistic_feat = paddle.expand(
+                holistic_feat, shape=[bsz, seq_len, hf_c])
+            y = self.prediction(paddle.concat((y, attn_feat, holistic_feat), 2))
+        else:
+            y = self.prediction(attn_feat)
+        # bsz * (seq_len + 1) * num_classes
+        if self.train_mode:
+            y = self.pred_dropout(y)
+
+        return y
+
+    def forward_train(self, feat, out_enc, label, img_metas):
+        '''
+        img_metas: [label, valid_ratio]
+        '''
+        if img_metas is not None:
+            assert len(img_metas[0]) == feat.shape[0]
+
+        valid_ratios = None
+        if img_metas is not None and self.mask:
+            valid_ratios = img_metas[-1]
+
+        label = label.cuda()
+        lab_embedding = self.embedding(label)
+        # bsz * seq_len * emb_dim
+        out_enc = out_enc.unsqueeze(1)
+        # bsz * 1 * emb_dim
+        in_dec = paddle.concat((out_enc, lab_embedding), axis=1)
+        # bsz * (seq_len + 1) * C
+        out_dec = self._2d_attention(
+            in_dec, feat, out_enc, valid_ratios=valid_ratios)
+        # bsz * (seq_len + 1) * num_classes
+
+        return out_dec[:, 1:, :]  # bsz * seq_len * num_classes
+
+    def forward_test(self, feat, out_enc, img_metas):
+        if img_metas is not None:
+            assert len(img_metas[0]) == feat.shape[0]
+
+        valid_ratios = None
+        if img_metas is not None and self.mask:
+            valid_ratios = img_metas[-1]
+
+        seq_len = self.max_seq_len
+        bsz = feat.shape[0]
+        start_token = paddle.full(
+            (bsz, ), fill_value=self.start_idx, dtype='int64')
+        # bsz
+        start_token = self.embedding(start_token)
+        # bsz * emb_dim
+        emb_dim = start_token.shape[1]
+        start_token = start_token.unsqueeze(1)
+        start_token = paddle.expand(start_token, shape=[bsz, seq_len, emb_dim])
+        # bsz * seq_len * emb_dim
+        out_enc = out_enc.unsqueeze(1)
+        # bsz * 1 * emb_dim
+        decoder_input = paddle.concat((out_enc, start_token), axis=1)
+        # bsz * (seq_len + 1) * emb_dim
+
+        outputs = []
+        for i in range(1, seq_len + 1):
+            decoder_output = self._2d_attention(
+                decoder_input, feat, out_enc, valid_ratios=valid_ratios)
+            char_output = decoder_output[:, i, :]  # bsz * num_classes
+            char_output = F.softmax(char_output, -1)
+            outputs.append(char_output)
+            max_idx = paddle.argmax(char_output, axis=1, keepdim=False)
+            char_embedding = self.embedding(max_idx)  # bsz * emb_dim
+            if i < seq_len:
+                decoder_input[:, i + 1, :] = char_embedding
+
+        outputs = paddle.stack(outputs, 1)  # bsz * seq_len * num_classes
+
+        return outputs
+
+
+class SARHead(nn.Layer):
+    def __init__(self,
+                 out_channels,
+                 enc_bi_rnn=False,
+                 enc_drop_rnn=0.1,
+                 enc_gru=False,
+                 dec_bi_rnn=False,
+                 dec_drop_rnn=0.0,
+                 dec_gru=False,
+                 d_k=512,
+                 pred_dropout=0.1,
+                 max_text_length=30,
+                 pred_concat=True,
+                 **kwargs):
+        super(SARHead, self).__init__()
+
+        # encoder module
+        self.encoder = SAREncoder(
+            enc_bi_rnn=enc_bi_rnn, enc_drop_rnn=enc_drop_rnn, enc_gru=enc_gru)
+
+        # decoder module
+        self.decoder = ParallelSARDecoder(
+            out_channels=out_channels,
+            enc_bi_rnn=enc_bi_rnn,
+            dec_bi_rnn=dec_bi_rnn,
+            dec_drop_rnn=dec_drop_rnn,
+            dec_gru=dec_gru,
+            d_k=d_k,
+            pred_dropout=pred_dropout,
+            max_text_length=max_text_length,
+            pred_concat=pred_concat)
+
+    def forward(self, feat, targets=None):
+        '''
+        img_metas: [label, valid_ratio]
+        '''
+        holistic_feat = self.encoder(feat, targets)  # bsz c
+
+        if self.training:
+            label = targets[0]  # label
+            label = paddle.to_tensor(label, dtype='int64')
+            final_out = self.decoder(
+                feat, holistic_feat, label, img_metas=targets)
+        if not self.training:
+            final_out = self.decoder(
+                feat,
+                holistic_feat,
+                label=None,
+                img_metas=targets,
+                train_mode=False)
+            # (bsz, seq_len, num_classes)
+
+        return final_out

ppocr/postprocess/__init__.py

@@ -25,7 +25,7 @@ from .db_postprocess import DBPostProcess, DistillationDBPostProcess
 from .east_postprocess import EASTPostProcess
 from .sast_postprocess import SASTPostProcess
 from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, DistillationCTCLabelDecode, NRTRLabelDecode, \
-    TableLabelDecode
+    TableLabelDecode, SARLabelDecode
 from .cls_postprocess import ClsPostProcess
 from .pg_postprocess import PGPostProcess
 from .pse_postprocess import PSEPostProcess
@@ -35,8 +35,8 @@ def build_post_process(config, global_config=None):
     support_dict = [
         'DBPostProcess', 'PSEPostProcess', 'EASTPostProcess', 'SASTPostProcess',
         'CTCLabelDecode', 'AttnLabelDecode', 'ClsPostProcess', 'SRNLabelDecode',
-        'PGPostProcess', 'DistillationCTCLabelDecode', 'NRTRLabelDecode',
-        'TableLabelDecode', 'DistillationDBPostProcess'
+        'PGPostProcess', 'DistillationCTCLabelDecode', 'TableLabelDecode',
+        'DistillationDBPostProcess', 'NRTRLabelDecode', 'SARLabelDecode'
     ]
 
     config = copy.deepcopy(config)

ppocr/postprocess/rec_postprocess.py

@@ -15,6 +15,7 @@ import numpy as np
 import string
 import paddle
 from paddle.nn import functional as F
+import re
 
 
 class BaseRecLabelDecode(object):
@@ -165,21 +166,21 @@ class NRTRLabelDecode(BaseRecLabelDecode):
                  use_space_char=True,
                  **kwargs):
         super(NRTRLabelDecode, self).__init__(character_dict_path,
-                                             character_type, use_space_char)
+                                              character_type, use_space_char)
 
     def __call__(self, preds, label=None, *args, **kwargs):
         if preds.dtype == paddle.int64:
             if isinstance(preds, paddle.Tensor):
                 preds = preds.numpy()
-            if preds[0][0]==2:
-                preds_idx = preds[:,1:]
+            if preds[0][0] == 2:
+                preds_idx = preds[:, 1:]
             else:
                 preds_idx = preds
 
             text = self.decode(preds_idx)
             if label is None:
                 return text
-            label = self.decode(label[:,1:])
+            label = self.decode(label[:, 1:])
         else:
             if isinstance(preds, paddle.Tensor):
                 preds = preds.numpy()
@@ -188,13 +189,13 @@ class NRTRLabelDecode(BaseRecLabelDecode):
             text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
             if label is None:
                 return text
-            label = self.decode(label[:,1:])
+            label = self.decode(label[:, 1:])
         return text, label
 
     def add_special_char(self, dict_character):
-        dict_character = ['blank','<unk>','<s>','</s>'] + dict_character
+        dict_character = ['blank', '<unk>', '<s>', '</s>'] + dict_character
         return dict_character
-    
+
     def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
         """ convert text-index into text-label. """
         result_list = []
@@ -203,10 +204,11 @@ class NRTRLabelDecode(BaseRecLabelDecode):
             char_list = []
             conf_list = []
             for idx in range(len(text_index[batch_idx])):
-                if text_index[batch_idx][idx] == 3: # end
+                if text_index[batch_idx][idx] == 3:  # end
                     break
                 try:
-                    char_list.append(self.character[int(text_index[batch_idx][idx])])
+                    char_list.append(self.character[int(text_index[batch_idx][
+                        idx])])
                 except:
                     continue
                 if text_prob is not None:
@@ -218,7 +220,6 @@ class NRTRLabelDecode(BaseRecLabelDecode):
         return result_list
 
 
-
 class AttnLabelDecode(BaseRecLabelDecode):
     """ Convert between text-label and text-index """
 
@@ -256,7 +257,8 @@ class AttnLabelDecode(BaseRecLabelDecode):
                     if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
                             batch_idx][idx]:
                         continue
-                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                char_list.append(self.character[int(text_index[batch_idx][
+                    idx])])
                 if text_prob is not None:
                     conf_list.append(text_prob[batch_idx][idx])
                 else:
@@ -386,10 +388,9 @@ class SRNLabelDecode(BaseRecLabelDecode):
 class TableLabelDecode(object):
     """  """
 
-    def __init__(self,
-                 character_dict_path,
-                 **kwargs):
-        list_character, list_elem = self.load_char_elem_dict(character_dict_path)
+    def __init__(self, character_dict_path, **kwargs):
+        list_character, list_elem = self.load_char_elem_dict(
+            character_dict_path)
         list_character = self.add_special_char(list_character)
         list_elem = self.add_special_char(list_elem)
         self.dict_character = {}
@@ -408,7 +409,8 @@ class TableLabelDecode(object):
         list_elem = []
         with open(character_dict_path, "rb") as fin:
             lines = fin.readlines()
-            substr = lines[0].decode('utf-8').strip("\n").strip("\r\n").split("\t")
+            substr = lines[0].decode('utf-8').strip("\n").strip("\r\n").split(
+                "\t")
             character_num = int(substr[0])
             elem_num = int(substr[1])
             for cno in range(1, 1 + character_num):
@@ -428,14 +430,14 @@ class TableLabelDecode(object):
     def __call__(self, preds):
         structure_probs = preds['structure_probs']
         loc_preds = preds['loc_preds']
-        if isinstance(structure_probs,paddle.Tensor):
+        if isinstance(structure_probs, paddle.Tensor):
             structure_probs = structure_probs.numpy()
-        if isinstance(loc_preds,paddle.Tensor):
+        if isinstance(loc_preds, paddle.Tensor):
             loc_preds = loc_preds.numpy()
         structure_idx = structure_probs.argmax(axis=2)
         structure_probs = structure_probs.max(axis=2)
-        structure_str, structure_pos, result_score_list, result_elem_idx_list = self.decode(structure_idx,
-                                                                                            structure_probs, 'elem')
+        structure_str, structure_pos, result_score_list, result_elem_idx_list = self.decode(
+            structure_idx, structure_probs, 'elem')
         res_html_code_list = []
         res_loc_list = []
         batch_num = len(structure_str)
@@ -450,8 +452,13 @@ class TableLabelDecode(object):
             res_loc = np.array(res_loc)
             res_html_code_list.append(res_html_code)
             res_loc_list.append(res_loc)
-        return {'res_html_code': res_html_code_list, 'res_loc': res_loc_list, 'res_score_list': result_score_list,
-                'res_elem_idx_list': result_elem_idx_list,'structure_str_list':structure_str}
+        return {
+            'res_html_code': res_html_code_list,
+            'res_loc': res_loc_list,
+            'res_score_list': result_score_list,
+            'res_elem_idx_list': result_elem_idx_list,
+            'structure_str_list': structure_str
+        }
 
     def decode(self, text_index, structure_probs, char_or_elem):
         """convert text-label into text-index.
@@ -516,3 +523,82 @@ class TableLabelDecode(object):
             assert False, "Unsupport type %s in char_or_elem" \
                           % char_or_elem
         return idx
+
+
+class SARLabelDecode(BaseRecLabelDecode):
+    """ Convert between text-label and text-index """
+
+    def __init__(self,
+                 character_dict_path=None,
+                 character_type='ch',
+                 use_space_char=False,
+                 **kwargs):
+        super(SARLabelDecode, self).__init__(character_dict_path,
+                                             character_type, use_space_char)
+
+        self.rm_symbol = kwargs.get('rm_symbol', False)
+
+    def add_special_char(self, dict_character):
+        beg_end_str = "<BOS/EOS>"
+        unknown_str = "<UKN>"
+        padding_str = "<PAD>"
+        dict_character = dict_character + [unknown_str]
+        self.unknown_idx = len(dict_character) - 1
+        dict_character = dict_character + [beg_end_str]
+        self.start_idx = len(dict_character) - 1
+        self.end_idx = len(dict_character) - 1
+        dict_character = dict_character + [padding_str]
+        self.padding_idx = len(dict_character) - 1
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """ convert text-index into text-label. """
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if int(text_index[batch_idx][idx]) == int(self.end_idx):
+                    if text_prob is None and idx == 0:
+                        continue
+                    else:
+                        break
+                if is_remove_duplicate:
+                    # only for predict
+                    if idx > 0 and text_index[batch_idx][idx - 1] == text_index[
+                            batch_idx][idx]:
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][
+                    idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = ''.join(char_list)
+            if self.rm_symbol:
+                comp = re.compile('[^A-Z^a-z^0-9^\u4e00-\u9fa5]')
+                text = text.lower()
+                text = comp.sub('', text)
+            result_list.append((text, np.mean(conf_list)))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+
+        if label is None:
+            return text
+        label = self.decode(label, is_remove_duplicate=False)
+        return text, label
+
+    def get_ignored_tokens(self):
+        return [self.padding_idx]

ppocr/utils/save_load.py

@@ -108,14 +108,15 @@ def load_dygraph_params(config, model, logger, optimizer):
         for k1, k2 in zip(state_dict.keys(), params.keys()):
             if list(state_dict[k1].shape) == list(params[k2].shape):
                 new_state_dict[k1] = params[k2]
-        else:
-            logger.info(
-                f"The shape of model params {k1} {state_dict[k1].shape} not matched with loaded params {k2} {params[k2].shape} !"
-            )
+            else:
+                logger.info(
+                    f"The shape of model params {k1} {state_dict[k1].shape} not matched with loaded params {k2} {params[k2].shape} !"
+                )
         model.set_state_dict(new_state_dict)
         logger.info(f"loaded pretrained_model successful from {pm}")
         return {}
 
+
 def load_pretrained_params(model, path):
     if path is None:
         return False
@@ -138,6 +139,7 @@ def load_pretrained_params(model, path):
     print(f"load pretrain successful from {path}")
     return model
 
+
 def save_model(model,
                optimizer,
                model_path,

tests/configs/det_mv3_db.yml

+Global:
+  use_gpu: false
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 1
+  save_model_dir: ./output/db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 400]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+    disable_se: True
+  Neck:
+    name: DBFPN
+    out_channels: 96
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5 
+  beta: 10 
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam #Momentum
+  #momentum: 0.9
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - Resize:
+          # size: [640, 640]
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 0
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 0
+    use_shared_memory: False

tests/configs/det_r50_vd_db.yml

+Global:
+  use_gpu: false
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 1
+  save_model_dir: ./output/db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 400]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: ResNet  #MobileNetV3
+    layers: 50
+  Neck:
+    name: DBFPN
+    out_channels: 256
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5 #5
+  beta: 10 #10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam #Momentum
+  #momentum: 0.9
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - Resize:
+          # size: [640, 640]
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 0
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 0
+    use_shared_memory: False

tests/configs/rec_icdar15_r34_train.yml

+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/ic15/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir: ./
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/en_dict.txt
+  character_type: EN
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_ic15.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False

tests/ocr_det_params.txt

@@ -13,34 +13,34 @@ train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
 null:null
 ##
 trainer:norm_train|pact_train
-norm_train:tools/train.py -c configs/det/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
-pact_train:deploy/slim/quantization/quant.py -c configs/det/det_mv3_db.yml -o
-fpgm_train:deploy/slim/prune/sensitivity_anal.py -c configs/det/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
+norm_train:tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:deploy/slim/quantization/quant.py -c tests/configs/det_mv3_db.yml -o
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
 distill_train:null
 null:null
 null:null
 ##
 ===========================eval_params=========================== 
-eval:tools/eval.py -c configs/det/det_mv3_db.yml -o
+eval:tools/eval.py -c tests/configs/det_mv3_db.yml -o
 null:null
 ##
 ===========================infer_params===========================
 Global.save_inference_dir:./output/
 Global.pretrained_model:
-norm_export:tools/export_model.py -c configs/det/det_mv3_db.yml -o 
-quant_export:deploy/slim/quantization/export_model.py -c configs/det/det_mv3_db.yml -o 
-fpgm_export:deploy/slim/prune/export_prune_model.py -c configs/det/det_mv3_db.yml -o 
+norm_export:tools/export_model.py -c tests/configs/det_mv3_db.yml -o 
+quant_export:deploy/slim/quantization/export_model.py -c tests/configs/det_mv3_db.yml -o 
+fpgm_export:deploy/slim/prune/export_prune_model.py -c tests/configs/det_mv3_db.yml -o 
 distill_export:null
 export1:null
 export2:null
 ##
-infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
-infer_export:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/det_mv3_db.yml -o
 infer_quant:False
 inference:tools/infer/predict_det.py
 --use_gpu:True|False
 --enable_mkldnn:True|False
---cpu_threads:1|6
+--cpu_threads:6
 --rec_batch_num:1
 --use_tensorrt:False|True
 --precision:fp32|fp16|int8
@@ -62,6 +62,21 @@ inference:./deploy/cpp_infer/build/ppocr det
 --precision:fp32|fp16
 --det_model_dir:
 --image_dir:./inference/ch_det_data_50/all-sum-510/
---save_log_path:null
+null:null
 --benchmark:True
+===========================serving_params===========================
+trans_model:-m paddle_serving_client.convert
+--dirname:./inference/ch_ppocr_mobile_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--serving_server:./deploy/pdserving/ppocr_det_mobile_2.0_serving/
+--serving_client:./deploy/pdserving/ppocr_det_mobile_2.0_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency=1
+op.det.local_service_conf.devices:null|0
+op.det.local_service_conf.use_mkldnn:True|False
+op.det.local_service_conf.thread_num:1|6
+op.det.local_service_conf.use_trt:False|True
+op.det.local_service_conf.precision:fp32|fp16|int8
+pipline:pipeline_http_client.py --image_dir=../../doc/imgs
 

tests/ocr_det_server_params.txt

@@ -13,7 +13,7 @@ train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
 null:null
 ##
 trainer:norm_train|pact_train
-norm_train:tools/train.py -c configs/det/det_r50_vd_db.yml -o Global.pretrained_model=""
+norm_train:tools/train.py -c tests/configs/det_r50_vd_db.yml -o Global.pretrained_model=""
 pact_train:null
 fpgm_train:null
 distill_train:null
@@ -21,13 +21,13 @@ null:null
 null:null
 ##
 ===========================eval_params=========================== 
-eval:tools/eval.py -c configs/det/det_mv3_db.yml -o
+eval:tools/eval.py -c tests/configs/det_r50_vd_db.yml -o
 null:null
 ##
 ===========================infer_params===========================
 Global.save_inference_dir:./output/
 Global.pretrained_model:
-norm_export:tools/export_model.py -c configs/det/det_r50_vd_db.yml -o 
+norm_export:tools/export_model.py -c tests/configs/det_r50_vd_db.yml -o 
 quant_export:null 
 fpgm_export:null
 distill_export:null

tests/ocr_ppocr_mobile_params.txt

+===========================train_params===========================
+model_name:ocr_system_mobile
+python:python3.7
+gpu_list:null
+Global.use_gpu:null
+Global.auto_cast:null
+Global.epoch_num:null
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:null
+Global.pretrained_model:null
+train_model_name:null
+train_infer_img_dir:null
+null:null
+##
+trainer:
+norm_train:null
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.pretrained_model:
+norm_export:null 
+quant_export:null 
+fpgm_export:null 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_system.py
+--use_gpu:True
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--rec_model_dir:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+===========================cpp_infer_params===========================
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr system
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--benchmark:True
+

tests/ocr_ppocr_server_params.txt

+===========================train_params===========================
+model_name:ocr_system_server
+python:python3.7
+gpu_list:null
+Global.use_gpu:null
+Global.auto_cast:null
+Global.epoch_num:null
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:null
+Global.pretrained_model:null
+train_model_name:null
+train_infer_img_dir:null
+null:null
+##
+trainer:
+norm_train:null
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.pretrained_model:
+norm_export:null
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/ch_ppocr_server_v2.0_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_system.py
+--use_gpu:True
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--rec_model_dir:./inference/ch_ppocr_server_v2.0_rec_infer/
+===========================cpp_infer_params===========================
+use_opencv:True
+infer_model:./inference/ch_ppocr_server_v2.0_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr system
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_ppocr_server_v2.0_rec_infer/
+--benchmark:True
\ No newline at end of file

tests/ocr_rec_params.txt

 ===========================train_params===========================
 model_name:ocr_rec
 python:python3.7
-gpu_list:0|2,3
+gpu_list:0|0,1
 Global.use_gpu:True|True
 Global.auto_cast:null
 Global.epoch_num:lite_train_infer=2|whole_train_infer=300
@@ -9,7 +9,7 @@ Global.save_model_dir:./output/
 Train.loader.batch_size_per_card:lite_train_infer=128|whole_train_infer=128
 Global.pretrained_model:null
 train_model_name:latest
-train_infer_img_dir:./train_data/ic15_data/train
+train_infer_img_dir:./inference/rec_inference
 null:null
 ##
 trainer:norm_train|pact_train
@@ -41,7 +41,7 @@ inference:tools/infer/predict_rec.py
 --use_gpu:True|False
 --enable_mkldnn:True|False
 --cpu_threads:1|6
---rec_batch_num:1
+--rec_batch_num:1|6
 --use_tensorrt:True|False
 --precision:fp32|fp16|int8
 --rec_model_dir:
@@ -49,3 +49,33 @@ inference:tools/infer/predict_rec.py
 --save_log_path:./test/output/
 --benchmark:True
 null:null
+===========================cpp_infer_params===========================
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr rec
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+===========================serving_params===========================
+trans_model:-m paddle_serving_client.convert
+--dirname:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--serving_server:./deploy/pdserving/ppocr_rec_mobile_2.0_serving/
+--serving_client:./deploy/pdserving/ppocr_rec_mobile_2.0_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency=1
+op.rec.local_service_conf.devices:null|0
+op.rec.local_service_conf.use_mkldnn:True|False
+op.rec.local_service_conf.thread_num:1|6
+op.rec.local_service_conf.use_trt:False|True
+op.rec.local_service_conf.precision:fp32|fp16|int8
+pipline:pipeline_http_client.py --image_dir=../../doc/imgs_words_en
\ No newline at end of file

tests/ocr_rec_server_params.txt

+===========================train_params===========================
+model_name:ocr_server_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_infer=2|whole_train_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_infer=128|whole_train_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train|pact_train
+norm_train:tools/train.py -c tests/configs/rec_icdar15_r34_train.yml -o
+pact_train:deploy/slim/quantization/quant.py -c tests/configs/rec_icdar15_r34_train.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c tests/configs/rec_icdar15_r34_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.pretrained_model:
+norm_export:tools/export_model.py -c tests/configs/rec_icdar15_r34_train.yml -o
+quant_export:deploy/slim/quantization/export_model.py -c tests/configs/rec_icdar15_r34_train.yml -o
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/ch_ppocr_server_v2.0_rec_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1|6
+--use_tensorrt:True|False
+--precision:fp32|fp16|int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================cpp_infer_params===========================
+use_opencv:True
+infer_model:./inference/ch_ppocr_server_v2.0_rec_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr rec
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+===========================serving_params===========================
+trans_model:-m paddle_serving_client.convert
+--dirname:./inference/ch_ppocr_server_v2.0_rec_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--serving_server:./deploy/pdserving/ppocr_rec_server_2.0_serving/
+--serving_client:./deploy/pdserving/ppocr_rec_server_2.0_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency=1
+op.rec.local_service_conf.devices:null|0
+op.rec.local_service_conf.use_mkldnn:True|False
+op.rec.local_service_conf.thread_num:1|6
+op.rec.local_service_conf.use_trt:False|True
+op.rec.local_service_conf.precision:fp32|fp16|int8
+pipline:pipeline_http_client.py --image_dir=../../doc/imgs_words_en
\ No newline at end of file

tests/prepare.sh

 #!/bin/bash
 FILENAME=$1
-# MODE be one of ['lite_train_infer' 'whole_infer' 'whole_train_infer', 'infer', 'cpp_infer']
+
+# MODE be one of ['lite_train_infer' 'whole_infer' 'whole_train_infer', 'infer', 'cpp_infer', 'serving_infer']
 MODE=$2
 
 dataline=$(cat ${FILENAME})
@@ -40,11 +41,13 @@ if [ ${MODE} = "lite_train_infer" ];then
     rm -rf ./train_data/ic15_data
     wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar
     wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar # todo change to bcebos
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar
     wget -nc -P ./deploy/slim/prune https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/sen.pickle
     
     cd ./train_data/ && tar xf icdar2015_lite.tar && tar xf ic15_data.tar
     ln -s ./icdar2015_lite ./icdar2015
     cd ../
+    cd ./inference && tar xf rec_inference.tar && cd ../
 elif [ ${MODE} = "whole_train_infer" ];then
     wget -nc -P  ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x0_5_pretrained.pdparams
     rm -rf ./train_data/icdar2015
@@ -61,64 +64,113 @@ elif [ ${MODE} = "whole_infer" ];then
     cd ./train_data/ && tar xf icdar2015_infer.tar && tar xf ic15_data.tar
     ln -s ./icdar2015_infer ./icdar2015
     cd ../
-elif [ ${MODE} = "infer" ] || [ ${MODE} = "cpp_infer" ];then
+elif [ ${MODE} = "infer" ];then
     if [ ${model_name} = "ocr_det" ]; then
-        eval_model_name="ch_ppocr_mobile_v2.0_det_infer"
+        eval_model_name="ch_ppocr_mobile_v2.0_det_train"
         rm -rf ./train_data/icdar2015
         wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
-        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar
         cd ./inference && tar xf ${eval_model_name}.tar && tar xf ch_det_data_50.tar && cd ../
     elif [ ${model_name} = "ocr_server_det" ]; then
         wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar
         wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
         cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
-    else 
+    elif  [ ${model_name} = "ocr_system_mobile" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif  [ ${model_name} = "ocr_system_server" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ocr_rec" ]; then
         rm -rf ./train_data/ic15_data
         eval_model_name="ch_ppocr_mobile_v2.0_rec_infer"
-        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ocr_server_rec" ]; then
+        rm -rf ./train_data/ic15_data
+        eval_model_name="ch_ppocr_server_v2.0_rec_infer"
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf rec_inference.tar && cd ../
+    fi 
+elif [ ${MODE} = "cpp_infer" ];then
+    if [ ${model_name} = "ocr_det" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ocr_rec" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar
         wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
-        cd ./inference && tar xf ${eval_model_name}.tar && tar xf ic15_data.tar && cd ../
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf rec_inference.tar && cd ../
+    elif  [ ${model_name} = "ocr_system" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
     fi 
 fi
 
+if [ ${MODE} = "serving_infer" ];then
+    # prepare serving env
+    python_name=$(func_parser_value "${lines[2]}")
+    wget https://paddle-serving.bj.bcebos.com/chain/paddle_serving_server_gpu-0.0.0.post101-py3-none-any.whl
+    ${python_name} -m pip install install paddle_serving_server_gpu-0.0.0.post101-py3-none-any.whl
+    ${python_name} -m pip install paddle_serving_client==0.6.1
+    ${python_name} -m pip install paddle-serving-app==0.6.3
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar
+    cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_ppocr_server_v2.0_det_infer.tar cd ../
+fi
+
 if [ ${MODE} = "cpp_infer" ];then
     cd deploy/cpp_infer
     use_opencv=$(func_parser_value "${lines[52]}")
     if [ ${use_opencv} = "True" ]; then
-        echo "################### build opencv ###################"
-        rm -rf 3.4.7.tar.gz opencv-3.4.7/
-        wget https://github.com/opencv/opencv/archive/3.4.7.tar.gz
-        tar -xf 3.4.7.tar.gz
+        if [ -d "opencv-3.4.7/opencv3/" ] && [ $(md5sum opencv-3.4.7.tar.gz | awk -F ' ' '{print $1}') = "faa2b5950f8bee3f03118e600c74746a" ];then
+            echo "################### build opencv skipped ###################"
+        else
+            echo "################### build opencv ###################"
+            rm -rf opencv-3.4.7.tar.gz opencv-3.4.7/
+            wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/opencv-3.4.7.tar.gz
+            tar -xf opencv-3.4.7.tar.gz
 
-        cd opencv-3.4.7/
-        install_path=$(pwd)/opencv-3.4.7/opencv3
+            cd opencv-3.4.7/
+            install_path=$(pwd)/opencv3
 
-        rm -rf build
-        mkdir build
-        cd build
+            rm -rf build
+            mkdir build
+            cd build
 
-        cmake .. \
-            -DCMAKE_INSTALL_PREFIX=${install_path} \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DBUILD_SHARED_LIBS=OFF \
-            -DWITH_IPP=OFF \
-            -DBUILD_IPP_IW=OFF \
-            -DWITH_LAPACK=OFF \
-            -DWITH_EIGEN=OFF \
-            -DCMAKE_INSTALL_LIBDIR=lib64 \
-            -DWITH_ZLIB=ON \
-            -DBUILD_ZLIB=ON \
-            -DWITH_JPEG=ON \
-            -DBUILD_JPEG=ON \
-            -DWITH_PNG=ON \
-            -DBUILD_PNG=ON \
-            -DWITH_TIFF=ON \
-            -DBUILD_TIFF=ON
+            cmake .. \
+                -DCMAKE_INSTALL_PREFIX=${install_path} \
+                -DCMAKE_BUILD_TYPE=Release \
+                -DBUILD_SHARED_LIBS=OFF \
+                -DWITH_IPP=OFF \
+                -DBUILD_IPP_IW=OFF \
+                -DWITH_LAPACK=OFF \
+                -DWITH_EIGEN=OFF \
+                -DCMAKE_INSTALL_LIBDIR=lib64 \
+                -DWITH_ZLIB=ON \
+                -DBUILD_ZLIB=ON \
+                -DWITH_JPEG=ON \
+                -DBUILD_JPEG=ON \
+                -DWITH_PNG=ON \
+                -DBUILD_PNG=ON \
+                -DWITH_TIFF=ON \
+                -DBUILD_TIFF=ON
 
-        make -j
-        make install
-        cd ../
-        echo "################### build opencv finished ###################"
+            make -j
+            make install
+            cd ../
+            echo "################### build opencv finished ###################"
+        fi
     fi
 
 
@@ -149,4 +201,4 @@ if [ ${MODE} = "cpp_infer" ];then
 
     make -j
     echo "################### build PaddleOCR demo finished ###################"
-fi
+fi
\ No newline at end of file