Function decode_and_track

examples/main_video.cpp:239–273

}

static void decode_and_track(vapp_state& app, int fi) {
    app.frame = sam3_decode_video_frame(app.video_path, fi);
    if (app.frame.data.empty()) {
        snprintf(app.status, sizeof(app.status), "Failed to decode frame %d.", fi);
        app.playing = false;
        return;
    }
    app.frame_index = fi;

    if (app.tracker_created) {
        if (app.visual_only)
            app.result = sam3_propagate_frame(*app.tracker, *app.state, *app.model, app.frame);
        else
            app.result = sam3_track_frame(*app.tracker, *app.state, *app.model, app.frame);
        app.frame_encoded = true;

        // Record instance presence in timeline
        if (fi >= (int)app.timeline.size())
            app.timeline.resize(fi + 1);
        app.timeline[fi].instances.clear();
        for (const auto& det : app.result.detections)
            app.timeline[fi].instances.push_back({det.instance_id, det.score});
        if (fi > app.timeline_max_frame) app.timeline_max_frame = fi;

        snprintf(app.status, sizeof(app.status), "Frame %d/%d — %d objects tracked",
                 fi, app.video_info.n_frames, (int)app.result.detections.size());
    } else {
        // Just encode and show the frame without tracking
        sam3_encode_image(*app.state, *app.model, app.frame);
        app.frame_encoded = true;
        app.result = {};
        snprintf(app.status, sizeof(app.status), "Frame %d/%d — no tracker active",
                 fi, app.video_info.n_frames);
    }
}

// Check if a click position lands on any tracked instance mask.
// Returns the instance_id or -1.