Introduction

In several video codecs, H264 and HEVC in particular, you may have one or more areas delimited by a region which corresponds to the area of interest that should be displayed.

In addition to the various level of cropping, there could be an additional container cropping.

Current limitations

The current approach, at least for H264, checks if only container cropping is present or if only bitstream cropping is, and applies it to the decoded frame. This approach mixes lavc and lavf usage and does not take into consideration the (sporadic and unfortunate) case where both container and bitstream cropping are present.

HWAccell testing is impacted by this, since it's not possible to offset the pixel on an opaque surface (which happens on left and top cropping).

There are multiple global flags that interfere with it in some way, such as unaligned or ingorecrop.

Proposed solutions

First of all current behavior should be preserved, possibly with an codec-private option set by default.

Secondly it is proposed that all cropping values should be exported and the application should decide what to do with those, on a full resolution frame.

An AVPacket side data should be added indicating:

An application like avprobe is expected to apply each rectangle separately and report each dimension.

Alternatively it could be possible to specify just a delta on each size, rather than the full size of each rectangle.


CategoryBlueprint