add Baidu warpctc option to reproduce CTC results of our paper.

README.md

@@ -11,6 +11,7 @@ Based on this framework, we recorded the 1st place of [ICDAR2013 focused scene t
 The difference between our paper and ICDAR challenge is summarized [here](https://github.com/clovaai/deep-text-recognition-benchmark/issues/13).
 
 ## Updates
+**Aug 3, 2020**: added [guideline to use Baidu warpctc](https://github.com/clovaai/deep-text-recognition-benchmark/pull/209) which reproduces CTC results of our paper. <br>
 **Dec 27, 2019**: added [FLOPS](https://github.com/clovaai/deep-text-recognition-benchmark/issues/125) in our paper, and minor updates such as log_dataset.txt and [ICDAR2019-NormalizedED](https://github.com/clovaai/deep-text-recognition-benchmark/blob/86451088248e0490ff8b5f74d33f7d014f6c249a/test.py#L139-L165). <br>
 **Oct 22, 2019**: added [confidence score](https://github.com/clovaai/deep-text-recognition-benchmark/issues/82), and arranged the output form of training logs. <br>
 **Jul 31, 2019**: The paper is accepted at International Conference on Computer Vision (ICCV), Seoul 2019, as an oral talk. <br>

test.py

@@ -23,6 +23,10 @@ def benchmark_all_eval(model, criterion, converter, opt, calculate_infer_time=Fa
     eval_data_list = ['IIIT5k_3000', 'SVT', 'IC03_860', 'IC03_867', 'IC13_857',
                       'IC13_1015', 'IC15_1811', 'IC15_2077', 'SVTP', 'CUTE80']
 
+    # # To easily compute the total accuracy of our paper.
+    # eval_data_list = ['IIIT5k_3000', 'SVT', 'IC03_867', 
+    #                   'IC13_1015', 'IC15_2077', 'SVTP', 'CUTE80']
+
     if calculate_infer_time:
         evaluation_batch_size = 1  # batch_size should be 1 to calculate the GPU inference time per image.
     else:
@@ -100,10 +104,17 @@ def validation(model, criterion, evaluation_loader, converter, opt):
             # Calculate evaluation loss for CTC deocder.
             preds_size = torch.IntTensor([preds.size(1)] * batch_size)
             # permute 'preds' to use CTCloss format
-            cost = criterion(preds.log_softmax(2).permute(1, 0, 2), text_for_loss, preds_size, length_for_loss)
+            if opt.baiduCTC:
+                cost = criterion(preds.permute(1, 0, 2), text_for_loss, preds_size, length_for_loss) / batch_size
+            else:
+                cost = criterion(preds.log_softmax(2).permute(1, 0, 2), text_for_loss, preds_size, length_for_loss)
 
             # Select max probabilty (greedy decoding) then decode index to character
-            _, preds_index = preds.max(2)
+            if opt.baiduCTC:
+                _, preds_index = preds.max(2)
+                preds_index = preds_index.view(-1)
+            else:
+                _, preds_index = preds.max(2)
             preds_str = converter.decode(preds_index.data, preds_size.data)
 
         else:
@@ -246,6 +257,7 @@ def test(opt):
     parser.add_argument('--sensitive', action='store_true', help='for sensitive character mode')
     parser.add_argument('--PAD', action='store_true', help='whether to keep ratio then pad for image resize')
     parser.add_argument('--data_filtering_off', action='store_true', help='for data_filtering_off mode')
+    parser.add_argument('--baiduCTC', action='store_true', help='for data_filtering_off mode')
     """ Model Architecture """
     parser.add_argument('--Transformation', type=str, required=True, help='Transformation stage. None|TPS')
     parser.add_argument('--FeatureExtraction', type=str, required=True, help='FeatureExtraction stage. VGG|RCNN|ResNet')

train.py

@@ -12,7 +12,7 @@
 import torch.utils.data
 import numpy as np
 
-from utils import CTCLabelConverter, AttnLabelConverter, Averager
+from utils import CTCLabelConverter, CTCLabelConverterForBaiduWarpctc, AttnLabelConverter, Averager
 from dataset import hierarchical_dataset, AlignCollate, Batch_Balanced_Dataset
 from model import Model
 from test import validation
@@ -45,7 +45,10 @@ def train(opt):
 
     """ model configuration """
     if 'CTC' in opt.Prediction:
-        converter = CTCLabelConverter(opt.character)
+        if opt.baiduCTC:
+            converter = CTCLabelConverterForBaiduWarpctc(opt.character)
+        else:
+            converter = CTCLabelConverter(opt.character)
     else:
         converter = AttnLabelConverter(opt.character)
     opt.num_class = len(converter.character)
@@ -86,7 +89,12 @@ def train(opt):
 
     """ setup loss """
     if 'CTC' in opt.Prediction:
-        criterion = torch.nn.CTCLoss(zero_infinity=True).to(device)
+        if opt.baiduCTC:
+            # need to install warpctc. see our guideline.
+            from warpctc_pytorch import CTCLoss 
+            criterion = CTCLoss()
+        else:
+            criterion = torch.nn.CTCLoss(zero_infinity=True).to(device)
     else:
         criterion = torch.nn.CrossEntropyLoss(ignore_index=0).to(device)  # ignore [GO] token = ignore index 0
     # loss averager
@@ -144,8 +152,12 @@ def train(opt):
         if 'CTC' in opt.Prediction:
             preds = model(image, text)
             preds_size = torch.IntTensor([preds.size(1)] * batch_size)
-            preds = preds.log_softmax(2).permute(1, 0, 2)
-            cost = criterion(preds, text, preds_size, length)
+            if opt.baiduCTC:
+                preds = preds.permute(1, 0, 2)  # to use CTCLoss format
+                cost = criterion(preds, text, preds_size, length) / batch_size
+            else:
+                preds = preds.log_softmax(2).permute(1, 0, 2)
+                cost = criterion(preds, text, preds_size, length)
 
         else:
             preds = model(image, text[:, :-1])  # align with Attention.forward
@@ -232,6 +244,7 @@ def train(opt):
     parser.add_argument('--rho', type=float, default=0.95, help='decay rate rho for Adadelta. default=0.95')
     parser.add_argument('--eps', type=float, default=1e-8, help='eps for Adadelta. default=1e-8')
     parser.add_argument('--grad_clip', type=float, default=5, help='gradient clipping value. default=5')
+    parser.add_argument('--baiduCTC', action='store_true', help='for data_filtering_off mode')
     """ Data processing """
     parser.add_argument('--select_data', type=str, default='MJ-ST',
                         help='select training data (default is MJ-ST, which means MJ and ST used as training data)')

utils.py

@@ -52,6 +52,53 @@ def decode(self, text_index, length):
         return texts
 
 
+class CTCLabelConverterForBaiduWarpctc(object):
+    """ Convert between text-label and text-index for baidu warpctc """
+
+    def __init__(self, character):
+        # character (str): set of the possible characters.
+        dict_character = list(character)
+
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            # NOTE: 0 is reserved for 'CTCblank' token required by CTCLoss
+            self.dict[char] = i + 1
+
+        self.character = ['[CTCblank]'] + dict_character  # dummy '[CTCblank]' token for CTCLoss (index 0)
+
+    def encode(self, text, batch_max_length=25):
+        """convert text-label into text-index.
+        input:
+            text: text labels of each image. [batch_size]
+        output:
+            text: concatenated text index for CTCLoss.
+                    [sum(text_lengths)] = [text_index_0 + text_index_1 + ... + text_index_(n - 1)]
+            length: length of each text. [batch_size]
+        """
+        length = [len(s) for s in text]
+        text = ''.join(text)
+        text = [self.dict[char] for char in text]
+
+        return (torch.IntTensor(text), torch.IntTensor(length))
+
+    def decode(self, text_index, length):
+        """ convert text-index into text-label. """
+        texts = []
+        index = 0
+        for l in length:
+            t = text_index[index:index + l]
+
+            char_list = []
+            for i in range(l):
+                if t[i] != 0 and (not (i > 0 and t[i - 1] == t[i])):  # removing repeated characters and blank.
+                    char_list.append(self.character[t[i]])
+            text = ''.join(char_list)
+
+            texts.append(text)
+            index += l
+        return texts
+
+
 class AttnLabelConverter(object):
     """ Convert between text-label and text-index """