[Feature request] Navigation and previewing

[subof]

Describe the feature

1/ In order to select a section for subtitle recognition, it is necessary to have a smooth scroll through the video. Currently, this is quite difficult even on RTX 3090Ti (HP Z ws), and it is challenging to find specific fragments if they are distributed in only a few locations.

2/ Additionally, it would be beneficial to have a preview button during the program's operation, allowing users to observe the correct placement of the area.

3/Furthermore, it would be advantageous to incorporate a quick scan feature for automatically detecting the subtitle position.

Thanks

[timminator]

Hi!

1.) You can use your left and right arrow keys to jump forwards and backwards. In the advanced settings you can define under keyboard seek step how granularly you want it to be. That should do what you want I think?

2.) What do you mean by that? You see the red crop box around the text you specify? That is your preview that is used for processing and OCR?

3.) This would be highly error prone due to text in logos or all kind of locations in a frame. The engine detects ALL. A simple crop box defined by the user is the savest solution in my opinion.

[timminator]

Did I answer your questions?

[subof]

In the first case, I meant manual rewinding through the video sequence in order to quickly scan arbitrary fragments by holding the rewind indicator with the mouse.

In the second case, I meant visual observation of the recognition position, as happens in video encoders.

[timminator]

Sorry for the late answer.
For 1.) I made a few changes. I improved the speed of the video preview, you can hold down the left and right button now and scroll a lot more smoothly through the video with the next update. But that's the best I can do I am afraid. The framework does not actually show the video but extracted frames encoded as a Picture. This takes processing and is also the only thing possible.
But you can already hold down the keys in the current version - depending on the resolution it's already smooth or a little bit less. I would change the seek step for this in the advanced settings to a higher value though. But this will be way better in the next version.

For 2.) Here I am still confused. You want to see the extracted images in Step 1 while processing them in Step 2? Or do you want to visually see where the OCR engine detected text? The latter one is also possible but doing this for every extracted frame would be way to computationally expensive. So I don't really consider this. Displaying the frame we extracted in Step 1 is doable while we are in Step 2 performing OCR on them, at least every few frames we could show one as this could also put a too high load on the system in addition to the OCR process.
Could you elaborate on this point?