Test sequential and random frames

Author: Dan-Flores

benchmarks/decoders/benchmark_decoders.py

@@ -136,7 +136,6 @@ def main() -> None:
 
     if args.verify_outputs:
         verify_outputs(decoders_to_run, video_paths, num_uniform_samples)
-        return
 
     results = run_benchmarks(
         decoders_to_run,

benchmarks/decoders/benchmark_decoders_library.py

@@ -1031,12 +1031,18 @@ def verify_outputs(decoders_to_run, video_paths, num_samples):
     # Import library to show frames that don't match
     from tensorcat import tensorcat
 
-    # Create TorchCodecPublic decoder to use as a baseline
-    torchcodec_display_name = decoder_registry["torchcodec_public"].display_name
-    options = decoder_registry["torchcodec_public"].default_options
-    kind = decoder_registry["torchcodec_public"].kind
-    torchcodec_public_decoder = kind(**options)
-
+    # Reuse TorchCodecPublic decoder with options, if provided, as the baseline
+    if torchcodec_display_name := next(
+        (name for name in decoders_to_run if "TorchCodecPublic" in name),
+        None,
+    ):
+        torchcodec_public_decoder = decoders_to_run[torchcodec_display_name]
+    # Create default TorchCodecPublic decoder to use as a baseline
+    else:
+        torchcodec_display_name = decoder_registry["torchcodec_public"].display_name
+        options = decoder_registry["torchcodec_public"].default_options
+        kind = decoder_registry["torchcodec_public"].kind
+        torchcodec_public_decoder = kind(**options)
     # Get frames using each decoder
     for video_file_path in video_paths:
         metadata = get_metadata(video_file_path)