strexp / model_srn.py
markytools's picture
added strexp
d61b9c7
raw
history blame
14.4 kB
"""
Copyright (c) 2019-present NAVER Corp.
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 torch
import torch.nn as nn
import torch.nn.functional as F
from modules_srn.transformation import TPS_SpatialTransformerNetwork
from modules_srn.feature_extraction import VGG_FeatureExtractor, RCNN_FeatureExtractor, ResNet_FeatureExtractor
from modules_srn.sequence_modeling import BidirectionalLSTM
from modules_srn.prediction import Attention
from modules_srn.resnet_aster import ResNet_ASTER
from modules_srn.bert import Bert_Ocr
from modules_srn.bert import Config
from modules_srn.SRN_modules import Transforme_Encoder, SRN_Decoder, Torch_transformer_encoder
from modules_srn.resnet_fpn import ResNet_FPN
import settings
import sys
# singleChar - if -1 then STRScore outputs all char, however if
# 0 - N, then it will output the single character confidence of the index 0 to N
class STRScore(nn.Module):
def __init__(self, opt, converter, device, gtStr="", enableSingleCharAttrAve=False, model=None):
super(STRScore, self).__init__()
self.enableSingleCharAttrAve = enableSingleCharAttrAve
self.singleChar = -1
self.opt = opt
self.converter = converter
self.device = device
self.gtStr = gtStr
self.model = model # Pass here if you want to use it
self.blank = torch.tensor([-1], dtype=torch.float).to(self.device)
self.separator = torch.tensor([-2], dtype=torch.float).to(self.device)
# singleChar - if >=0, then the output of STRScore will only be a single character
# instead of a whole. The char index will be equal to the parameter "singleChar".
def setSingleCharOutput(self, singleChar):
self.singleChar = singleChar
def forward(self, preds):
preds = preds[2] # Access second index
bs = preds.shape[0]
# text_for_loss, length_for_loss = self.converter.encode(labels, batch_max_length=self.opt.batch_max_length)
text_for_loss_length = self.opt.batch_max_length + 1
# _, preds_index = preds.topk(1, dim=-1, largest=True, sorted=True)
# preds_index = preds_index.view(-1, self.converter.batch_max_length)
# print("preds shape: ", preds.shape)
# print("preds_index: ", preds_index)
# preds_str = self.converter.decode(preds_index, length_for_pred)
if settings.MODEL == 'vitstr':
preds_str = self.converter.decode(preds_index[:, 1:], length_for_pred)
elif settings.MODEL == 'srn':
_, preds_index = preds.max(2)
length_for_pred = torch.IntTensor([self.opt.batch_max_length] * bs).to(self.device)
preds_str = self.converter.decode(preds_index, length_for_pred)
# sys.exit()
elif settings.MODEL == 'parseq':
preds_str, confidence = self.model.tokenizer.decode(preds)
# Confidence score
# ARGMAX calculation
sum = torch.FloatTensor([0]*bs).to(self.device)
if self.opt.confidence_mode == 0:
preds_prob = F.softmax(preds, dim=2)
# preds_prob shape: torch.Size([1, 25, 96])
preds_max_prob, preds_max_idx = preds_prob.max(dim=2)
# preds_max_prob shape: torch.Size([1, 25])
confidence_score_list = []
count = 0
for one_hot_preds, pred, pred_max_prob in zip(preds_prob, preds_str, preds_max_prob):
if settings.MODEL == 'vitstr' or settings.MODEL == 'srn':
if self.enableSingleCharAttrAve:
one_hot = one_hot_preds[self.singleChar, :]
# pred = pred[self.singleChar]
pred_max_prob = pred_max_prob[self.singleChar]
else:
pred_EOS = pred.find('[s]')
pred = pred[:pred_EOS]
pred_max_prob = pred_max_prob[:pred_EOS]
# if pred_max_prob.shape[0] == 0: continue
if self.enableSingleCharAttrAve:
sum = one_hot
# sum shape: torch.Size([96])
sum = sum.unsqueeze(0)
else:
if self.opt.scorer == "cumprod":
confidence_score = pred_max_prob.cumprod(dim=0)[-1] ### Maximum is 1
sum[count] += confidence_score
elif self.opt.scorer == "mean":
confidence_score = torch.mean(pred_max_prob) ### Maximum is 1
sum[count] += confidence_score
sum = sum.unsqueeze(1)
elif settings.MODEL == 'parseq':
if self.enableSingleCharAttrAve:
one_hot = one_hot_preds[self.singleChar, :]
# pred = pred[self.singleChar]
pred_max_prob = pred_max_prob[self.singleChar]
else:
pred_EOS = len(pred) # Predition string already has no EOS, fully intact
pred_max_prob = pred_max_prob[:pred_EOS]
# if pred_max_prob.shape[0] == 0: continue
if self.enableSingleCharAttrAve:
sum = one_hot
# sum shape: torch.Size([96])
sum = sum.unsqueeze(0)
else:
if self.opt.scorer == "cumprod":
confidence_score = pred_max_prob.cumprod(dim=0)[-1] ### Maximum is 1
sum[count] += confidence_score
elif self.opt.scorer == "mean":
confidence_score = torch.mean(pred_max_prob) ### Maximum is 1
sum[count] += confidence_score
sum = sum.unsqueeze(1)
count += 1
# return sum.detach().cpu().numpy()
# print("sumshape: ", sum.shape)
elif self.opt.confidence_mode == 1:
preds_prob = F.softmax(preds, dim=2)
### Predicted indices
preds_max_prob = torch.argmax(preds_prob, 2)
# print("preds_max_prob shape: ", preds_max_prob.shape)
### Ground truth indices
gtIndices, _ = self.converter.encode([self.gtStr for i in range(0,preds_prob.shape[0])], batch_max_length=self.opt.batch_max_length-1)
# print("gtIndices shape: ", gtIndices.shape)
### Acquire levenstein distance
m = torch.tensor([preds_prob.shape[1] for i in range(0, gtIndices.shape[0])], dtype=torch.float).to(self.device)
n = torch.tensor([preds_prob.shape[1] for i in range(0, gtIndices.shape[0])], dtype=torch.float).to(self.device)
# print("m: ", m)
# print("preds_max_prob dtype: ", preds_max_prob.dtype)
# print("gtIndices dtype: ", gtIndices.dtype)
preds_max_prob = preds_max_prob.type(torch.float)
gtIndices = gtIndices.type(torch.float)
r = levenshtein_distance(preds_max_prob.to(self.device), gtIndices.to(self.device), n, m, torch.cat([self.blank, self.separator]), torch.empty([], dtype=torch.float).to(self.device))
# print("r shape: ", r.shape)
# confidence_score_list = []
# count = 0
# for pred, pred_max_prob in zip(preds_str, preds_max_prob):
# if 'Attn' in self.opt.Prediction:
# pred_EOS = pred.find('[s]')
# pred = pred[:pred_EOS]
# pred_max_prob = pred_max_prob[:pred_EOS] ### Use score of all letters
# # pred_max_prob = pred_max_prob[0:1] ### Use score of first letter only
# if pred_max_prob.shape[0] == 0: continue
# confidence_score = pred_max_prob.cumprod(dim=0)[-1]
# sum[count] += confidence_score
# count += 1
# return sum.detach().cpu().numpy()
# print("sumshape: ", sum.shape)
# sum = sum.unsqueeze(1)
rSoft = F.softmax(r[:,2].type(torch.float))
# rSoft = rSoft.contiguous()
rNorm = rSoft.max()-rSoft
sum = rNorm.unsqueeze(1)
return sum
class Model(nn.Module):
def __init__(self, opt):
super(Model, self).__init__()
self.opt = opt
self.stages = {'Trans': opt.Transformation, 'Feat': opt.FeatureExtraction,
'Seq': opt.SequenceModeling, 'Pred': opt.Prediction}
""" Transformation """
if opt.Transformation == 'TPS':
self.Transformation = TPS_SpatialTransformerNetwork(
F=opt.num_fiducial, I_size=(opt.imgH, opt.imgW), I_r_size=(opt.imgH, opt.imgW), I_channel_num=opt.input_channel)
else:
print('No Transformation module specified')
""" FeatureExtraction """
if opt.FeatureExtraction == 'VGG':
self.FeatureExtraction = VGG_FeatureExtractor(opt.input_channel, opt.output_channel)
elif opt.FeatureExtraction == 'RCNN':
self.FeatureExtraction = RCNN_FeatureExtractor(opt.input_channel, opt.output_channel)
elif opt.FeatureExtraction == 'ResNet':
self.FeatureExtraction = ResNet_FeatureExtractor(opt.input_channel, opt.output_channel)
self.AdaptiveAvgPool = nn.AdaptiveAvgPool2d((None, 1)) # Transform final (imgH/16-1) -> 1
elif opt.FeatureExtraction == 'AsterRes':
self.FeatureExtraction = ResNet_ASTER(opt.input_channel, opt.output_channel)
elif opt.FeatureExtraction == 'ResnetFpn':
self.FeatureExtraction = ResNet_FPN()
else:
raise Exception('No FeatureExtraction module specified')
self.FeatureExtraction_output = opt.output_channel # int(imgH/16-1) * 512
""" Sequence modeling"""
if opt.SequenceModeling == 'BiLSTM':
self.SequenceModeling = nn.Sequential(
BidirectionalLSTM(self.FeatureExtraction_output, opt.hidden_size, opt.hidden_size),
BidirectionalLSTM(opt.hidden_size, opt.hidden_size, opt.hidden_size))
self.SequenceModeling_output = opt.hidden_size
elif opt.SequenceModeling == 'Bert':
cfg = Config()
cfg.dim = opt.output_channel; cfg.dim_c = opt.output_channel # 降维减少计算量
cfg.p_dim = opt.position_dim # 一张图片cnn编码之后的特征序列长度
cfg.max_vocab_size = opt.batch_max_length + 1 # 一张图片中最多的文字个数, +1 for EOS
cfg.len_alphabet = opt.alphabet_size # 文字的类别个数
self.SequenceModeling = Bert_Ocr(cfg)
elif opt.SequenceModeling == 'SRN':
self.SequenceModeling = Transforme_Encoder(n_layers=2, n_position=opt.position_dim)
# self.SequenceModeling = Torch_transformer_encoder(n_layers=2, n_position=opt.position_dim)
self.SequenceModeling_output = 512
else:
print('No SequenceModeling module specified')
self.SequenceModeling_output = self.FeatureExtraction_output
""" Prediction """
if opt.Prediction == 'CTC':
self.Prediction = nn.Linear(self.SequenceModeling_output, opt.num_class)
elif opt.Prediction == 'Attn':
self.Prediction = Attention(self.SequenceModeling_output, opt.hidden_size, opt.num_class)
elif opt.Prediction == 'Bert_pred':
pass
elif opt.Prediction == 'SRN':
self.Prediction = SRN_Decoder(n_position=opt.position_dim, N_max_character=opt.batch_max_character + 1, n_class=opt.alphabet_size)
else:
raise Exception('Prediction is neither CTC or Attn')
def forward(self, input, text=None, is_train=True):
""" Transformation stage """
if not self.stages['Trans'] == "None":
input = self.Transformation(input)
""" Feature extraction stage """
visual_feature = self.FeatureExtraction(input)
# if self.stages['Feat'] == 'AsterRes' or self.stages['Feat'] == 'ResnetFpn':
if self.stages['Feat'] == 'AsterRes' or self.stages['Feat'] == 'ResnetFpn':
b, c, h, w = visual_feature.shape
visual_feature = visual_feature.permute(0, 1, 3, 2)
visual_feature = visual_feature.contiguous().view(b, c, -1)
visual_feature = visual_feature.permute(0, 2, 1) # batch, seq, feature
else:
visual_feature = self.AdaptiveAvgPool(visual_feature.permute(0, 3, 1, 2)) # [b, c, h, w] -> [b, w, c, h]
visual_feature = visual_feature.squeeze(3)
""" Sequence modeling stage """
if self.stages['Seq'] == 'BiLSTM':
contextual_feature = self.SequenceModeling(visual_feature)
elif self.stages['Seq'] == 'Bert':
pad_mask = text
contextual_feature = self.SequenceModeling(visual_feature, pad_mask)
elif self.stages['Seq'] == 'SRN':
contextual_feature = self.SequenceModeling(visual_feature, src_mask=None)[0]
else:
contextual_feature = visual_feature # for convenience. this is NOT contextually modeled by BiLSTM
""" Prediction stage """
if self.stages['Pred'] == 'CTC':
prediction = self.Prediction(contextual_feature.contiguous())
elif self.stages['Pred'] == 'Bert_pred':
prediction = contextual_feature
elif self.stages['Pred'] == 'SRN':
prediction = self.Prediction(contextual_feature)
else:
prediction = self.Prediction(contextual_feature.contiguous(), text, is_train, batch_max_length=self.opt.batch_max_length)
return prediction