- Generated by
1.9.8
|
liblloyal 1.0.0
Branched Inference for llama.cpp
|
The sampled frames of one video, in order, each an encoded image with the time it was sampled at. More...
#include <lloyal/mtmd.hpp>
Public Attributes | |
| std::span< const std::span< const uint8_t > > | frames |
| std::span< const int64_t > | timestamps_ms |
| int64_t | timestamp_interval_ms = 5000 |
| std::span< const VideoNote > | notes |
The sampled frames of one video, in order, each an encoded image with the time it was sampled at.
Borrowed, and consumed during construction.
Laid out as llama.cpp's video helper lays out a decoded clip: "Video:", every frame mergeable, and a "[XmY.YYs]" timestamp after the frame that crosses each timestamp_interval_ms (0 disables them). Intervals are counted from the first frame's time, so a window cut from a clip's middle is laid out as the clip's start is. A note joins the first timestamp at or after its time; notes past the last one close the video. Notes only join timestamp text the layout already has, so they add no text between frames and leave temporal fusion unchanged.
| std::span<const std::span<const uint8_t> > lloyal::VideoFrames::frames |
1.9.8