GUI "entire video" frame selection generates malformed --frames 0,-N, causing IndexError

[Taschimoz]

Description

Running inference/tracking from the GUI (Predict → Run Inference) with the frame
range set to "entire video" generates a malformed --frames 0,-N argument. It
appears to be parsed as a two-element frame list — frame 0 and frame -N —
rather than the range 0 to N. The negative value resolves to a frame index at
the end of the video, and the run fails with IndexError: Failed to read frame index N.

Selecting "entire video" reproducibly produces --frames 0,-N, where N tracks the
video length (observed 0,-10245, 0,-10240, 0,-10239 across different videos).

Environment

• sleap-nn 0.2.0
• Python 3.13.12
• PyTorch 2.10.0+cu129, CUDA 12.9
• OS: Linux (Rocky 8, HPC cluster)

Steps to reproduce

1. Open a project with a video in the GUI.
2. Predict → Run Inference; configure models + tracking.
3. Set the frame range to "entire video".
4. Run.

Expected

Inference runs over all frames of the video.

Actual

The GUI builds a command containing --frames 0,-10245, interpreted as the frame
list [0, -10245]; the negative index maps to frame 10245 and fails.

Generated command (paths redacted):sleap track --gui --data_path /path/to/project.slp --video_index 0
--frames 0,-10245
--model_paths /path/to/centroid_model
--model_paths /path/to/centered_instance_model
-o /path/to/output.predictions.slp
--max_instances 2 --tracking --track_matching_method hungarian
--tracking_window_size 15 --use_flow --post_connect_single_breaks
--features keypoints --scoring_method oks
Error Message:
File ".../sleap_nn/inference/predictors.py", line 1541, in _make_labeled_frames_from_generator
untracked_instances=inst, frame_idx=frame_idx, image=lf.image
File ".../sleap_io/model/labeled_frame.py", line 360, in image
return self.video[self.frame_idx]
File ".../sleap_io/io/video_reading.py", line 840, in _read_frame
raise IndexError(f"Failed to read frame index {frame_idx}.")
IndexError: Failed to read frame index 10245.
Error: Inference failed with code 1
Would love it if i could get some help in running inference over all videos and if this bug can be fixed. Thanks!

[tom21100227]

@Taschimoz Thank you for catching this bug Kerry!

This is a quirk from a mismatch in indexing methods (See #2790 ) I am guessing your video have 10245 frames in total. CLI uses 0-indexing, so it goes 0th frame, 1th frame, and your last frame would be the 10244th frame. On the GUI it shows up at X/10245 frames because there are in total 10245 frames.

There must've been some misconfiguration happening on the GUI side, I am looking into this right now.

Best,
Tom

[tom21100227]

@Taschimoz ,

I believe I figured out the error and fixed it (will be out in the next release). In the mean time, try the fix I described above (or just omit the --frames flag, sleap-nn would inference the entire video when you don't specify it). If it still does not work let us know and we'll take a further look.

Best,
Tom