Recognition Confidence metric

[daviembrito]

Hi!

Does someone know what metric and formula is used to calcule recognition confidence for character recognition? Didn't find anything around here. Thanks!

[SimWalther]

Hi, I'm interested in knowing too. I searched in the documentation and I couldn't find anything.
I looked at the PP-OCRv3 (PP-OCRv3: More Attempts for the Improvement of Ultra Lightweight OCR System) paper and I couldn't find mention of confidence but I found out that the text recognition model (SVTR-LCNet) make use of the CTC loss but guided by an attention mechanism (GTC).

The CTC paper talks about the total probability of the label sequences:
"A CTC network has a softmax output layer (Bridle, 1990) with one more unit than there are labels in L. The activations of the first |L| units are interpreted as the probabilities of observing the corresponding labels at particular times. The activation of the extra unit is the probability of observing a ‘blank’, or no label. Together, these outputs define the probabilities of all possible ways of aligning all possible label sequences with the input sequence. The total probability of any one label sequence can then be found by summing the probabilities of its different alignments." (Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks)

I also found an example but with RCNN where the CTC output probability is used as the confidence (Dive into OCR Text Recognition Practice):

"
When the network is randomly initialized, the output results are unordered. Through SoftMax, the prediction results with the highest probability at each time step can be obtained, where: pred_id represents the predicted tag ID, and pre_scores represents the confidence of the predicted result:

ctc_head = CTCHead(in_channels=96, out_channels=37)
predict = ctc_head(sequence)
print("predict shape:", predict.shape)
result = F.softmax(predict, axis=2)
pred_id = paddle.argmax(result, axis=2)
pred_socres = paddle.max(result, axis=2)
print("pred_id:", pred_id)
print("pred_scores:", pred_socres)

"

So I think that the confidence is the output label sequences softmax probabilities but I may be wrong.