Update load_video node for more accurate color conversion and chroma …

…reconstruction (#2532)

* Update load_video.py for more accurate color conversion and chroma reconstruction

This PR updates the ffmpeg based load_video node to use a more accurate color decoding and chroma reconstruction path.  

This node now uses lanczos filtering by default and also skips the previously present extra conversion from rgb24 to bgr24 by having ffmpeg send a bgr24 rawvideo pipe natively to chaiNNer

* Update load_video.py

remove cv2 import as this is no longer needed.

* Format with ruff

---------

Co-authored-by: Joey Ballentine <[email protected]>

backend/src/packages/chaiNNer_standard/image/video_frames/load_video.py

@@ -4,7 +4,6 @@
 from pathlib import Path
 from typing import Any
 
-import cv2
 import ffmpeg
 import numpy as np
 
@@ -68,7 +67,12 @@ def load_video_node(
 
     ffmpeg_reader = (
         ffmpeg.input(path)
-        .output("pipe:", format="rawvideo", pix_fmt="rgb24")
+        .output(
+            "pipe:",
+            format="rawvideo",
+            pix_fmt="bgr24",
+            sws_flags="lanczos+accurate_rnd+full_chroma_int+full_chroma_inp+bitexact",
+        )
         .run_async(pipe_stdout=True, cmd=ffmpeg_path)
     )
 
@@ -123,7 +127,6 @@ def iterator():
                 print("Can't receive frame (stream end?). Exiting ...")
                 break
             in_frame = np.frombuffer(in_bytes, np.uint8).reshape([height, width, 3])
-            in_frame = cv2.cvtColor(in_frame, cv2.COLOR_RGB2BGR)
             yield in_frame, index
             index += 1