Spaces:
Running
Resutls are good but still I need help
For me results are somewhat 70% accurate but could you please guide on how to achieve better accuracy?
Sorry to note that this model is mainly trained on Chinese data. However, as I just checked your example, except for one or two missing boxes due to small characters and blurs, the recognition results are mostly correct?
ok, any chances to capture all arrows ?? I can see only 1-2 arrows getting captured
The performance of the demo is bounded by the detection module. Now in order to achieve high coverage, there are a number of small boxes. In the next version, the merging of results in one line will be added, and now we are trying to find a good solution for that.
It is because the detection module just didn't detect some arrows as texts. Would you accept tuning parameters by yourself? I could consider expose some parameters for tuning the detection, like the threshold for probabilities of detection.
Also, I would like to get some suggestions. In what scenarios, do symbols like arrows matter to you? In terms of accuracy and efficiency, which matters most to you?
Hi Justin,
Yes I am open to tune parameters by myself. but I would need your guidance over the same.
To me, all arrows matter a lot and are very critical.
efficiency is not an issue but accuracy matters