JingsongLi opened a new pull request, #1371:
URL: https://github.com/apache/paimon-rust/pull/1371

   ### Purpose
   
   Let Python callers assemble ordered media rows into a frame-accurate MP4. 
This builds on the packed-video `moov` locator introduced in #1347.
   
   ### Brief change log
   
   - Add `SQLContext.read_video()` for ordered query results containing packed 
video frame descriptors, inline images, or image BLOB descriptors. Each row 
contributes one frame.
   - Add `video_clip()` and use the MP4 sample timeline to seek to the 
preceding keyframe for frame-accurate extraction.
   - Copy compressed AVC or HEVC packets when a native-timing range starts at 
an IDR and spans consecutive decode samples. Other ranges use the H.264 
encoding path.
   
   The implementation was AI-assisted. The indexed path checks that the MP4 
sample timeline matches the demuxed timestamps and packet positions. The remux 
path is limited to progressive, independently decodable ranges with a verified 
output timeline. HEVC midstream cuts require complete out-of-band parameter 
sets, and open-GOP pictures trigger frame-accurate encoding.
   
   ### Tests
   
   - `cargo fmt --all -- --check`
   - `git diff --check`
   - 143 tests passed across the Python DataFusion, SQL stream, BLOB consumer, 
and BLOB URI reader suites. Regression coverage includes B-frame order, 
variable frame timing, mixed media across batches, seek fallback, 
byte-identical AVC/HEVC packet copies, and open-GOP fallback.
   
   ### API and Format
   
   Adds the Python `SQLContext.read_video()` method and the `video_clip()` SQL 
helper. The packed `.video` and frame descriptor wire formats stay compatible 
with existing files.
   
   ### Documentation
   
   Adds SQL documentation for frame ranges, output timing, and remux 
eligibility.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to