def batch_cer_accuracy(preds, labels, label_lengths):
pred_sentences = get_most_probable(preds)
labels_list = labels.tolist()
idx = 0
cer = []
for i, length in enumerate(label_lengths.cpu().tolist()):
pred_sentence = pred_sentences[i]
gt_sentence = sequence_to_string(labels_list[idx:idx + length])
cer.append(cerCalc(pred_sentence, gt_sentence))
idx += length
return np.sum(cer)
def validate_update_function(engine, batch):
img, labels, label_lengths, image_lengths = batch
y_pred = model(img.to(device))
if np.random.rand() > 0.99:
pred_sentences = get_most_probable(y_pred)
labels_list = labels.tolist()
idx = 0
for i, length in enumerate(label_lengths.cpu().tolist()):
pred_sentence = pred_sentences[i]
gt_sentence = sequence_to_string(labels_list[idx:idx + length])
idx += length
print(f"Pred sentence: {pred_sentence}, GT: {gt_sentence}")
return (y_pred, labels, label_lengths)
if not data_enc:
return self.__getitem__(np.random.choice(range(len(self))))
data = msgpack.unpackb(data_enc, object_hook=m.decode, raw=False)
img = data['img']
label = data['label']
if self.transform is not None:
img = self.transform(img, self.epoch)
if self.target_transform is not None:
label = self.target_transform(label)
return (img, label)
if __name__ == "__main__":
from utils.config import lmdb_root_path
from datasets.librispeech import sequence_to_string
lmdb_commonvoice_root_path = "lmdb-databases-common_voice"
lmdb_airtel_root_path = "lmdb-databases-airtel"
trainCleanPath = os.path.join(lmdb_root_path, 'train-labelled')
trainOtherPath = os.path.join(lmdb_root_path, 'train-unlabelled')
trainCommonVoicePath = os.path.join(lmdb_commonvoice_root_path,
'train-labelled-en')
testAirtelPath = os.path.join(lmdb_airtel_root_path, 'test-labelled-en')
roots = [trainCleanPath, trainOtherPath, trainCommonVoicePath]
dataset = lmdbMultiDataset(roots=[testAirtelPath])
print(
sequence_to_string(dataset[np.random.choice(
len(dataset))][1].tolist()))