Jogging to position in a long video.

[ruck94301]

User reports that positioning a "cursor" in a long video ("jogging"?),
to locate a section to work on, can require a LOT of clicking. User
might make a macro to click 1000 times on their behalf, but that
shouldn't be necessary, right?

Does this mean a non-uniform jogging needs to be implemented? Or
is this a case where there exists a workaround or better approach
like "use ffmpeg to find and isolate the desired video segment."

On 2025-10-18, Henry asks for more details about the use case

Can you elaborate on this a little more, maybe with some examples? What do you want to achieve?

[LSBuenoJr]

I think my comment on issue #313 is also relevant here.

Colleagues are currently struggling to isolate behaviors of interest that are rare and occur later into the video. While re-exporting video segments outside of LabGym is a possible workaround, I believe there’s a more streamlined approach we could implement.

1. Run the usual pipeline for foreground extraction and animation / pattern-image generation (no change here).
2. Then, leverage the scrollable timeline feature discussed on issue In a display array of gifs, the gifs are ordered by filename instead of by time. #313 to navigate through the generated dataset (not the raw video) however you want.

[yujiahu415]

Using the sorting UI in LabGym is not the only way to sort behavior examples in LabGym. If the video is too long, first, you don't need to generate a pair of examples at each frame, instead, try use larger interval--you don't need to sort all the examples you generated either. You don't need that many examples. Second, if going over examples one by one in the sorting UI is too slow, try native folder system, in which you can view folder content as large icon and preview all the examples and choose which time window to focus on.

[ruck94301]

This entry is NOT responsive to Henry's suggestion of alternative approaches.
I'm simply copying in earlier related comments from @LSBuenoJr about the use case and imagined scrollable timeline feature.
As Henry has pointed out, (a) this might be accomplished with a plug-in instead of a new feature in LabGym proper, and (b) there are some downsides that worry him, wrt tuning with extra context info.

Lezio said:
...concerns the manual behavior sorter tool.

As users sort behaviors (animation and pattern-image pairs), the sorted items disappear from view. For example, if a user pauses and then resumes works days later, the remaining "behavior gifs" will be ordered by filename based on the still-unsorted set. This can dismantle ethologically meaningful behavioral sequences, making sequential labeling temporally disjointed.

Ideally, we’d like a scrollable timeline for the sorter tool that maintains the time axis of the original video, while preserving all current sorting functionalities. This would let users:

• Navigate by time (e.g., jump directly to, and label, rare behaviors that occur hours into the video).
• Resume labeling in any order (e.g., chronologically, from the end, or by skipping to notable events) while maintaining the sequential context of the behaviors for later adjustments/refinements.
• Always view behavior events in their original temporal order, regardless of sorting status.

Currently, sorted behaviors are moved to behavior-specific folders under-the-hood - I know that. However, it should be feasible to also preserve references or copies of the animation and pattern image pairs, specifically for rendering the scrollable timeline. Importantly, the scrollable timeline should include color-coded labels to reflect past or ongoing sorting efforts - similar to what Deep Ethogram does for individual video frames.