two questions about the detect result

[haooxia]

Hi, I have two questions:

First, may I ask why Figure 1 (inscription rubbing) is basically unrecognizable, but after binarization (Figure 2) the effect is very good? I guess it is because the background color and font color of the training set are different from Figure 1. (Figure 1) (Figure 2)

Second, Figure 3 shows the detection result of blocky noise. Whether every character can be identified for the blue part (9 characters are involved in the red bounding box in the figure), and whether there is a method to detect the green part in this case? (Figure 3)

[Tverous]

Hi @haoxia1,

Apologies for the delayed response.

Regarding Question 1, I believe you're right.

As for Question 2, you may utilize other language models, such as N-gram or Pointer-Generator Network, to forecast the subsequent word based on the given prior context.

[haooxia]

Hi @haoxia1,

Apologies for the delayed response.

Regarding Question 1, I believe you're right.

As for Question 2, you may utilize other language models, such as N-gram or Pointer-Generator Network, to forecast the subsequent word based on the given prior context.

Thanks so much for your reply, I will try it.