benchmark/batch_bench_record.md

Batch Processing Benchmark Results

Ablation Study

Detector	Batch	Cleaner	TorchCompile	Bf16	Time (s)	Speedup
YOLO	×	LAMA	×	×	44.33	-
YOLO	×	E2FGVI	×	×	142.42	1.00×
YOLO	×	E2FGVI	✓	×	117.19	1.22×
YOLO	4	E2FGVI	✓	×	82.63	1.72×
YOLO	4	E2FGVI	✓	✓	58.60	2.43×

Note:

• Speedup is calculated relative to the E2FGVI baseline (142.42s).
• LAMA is a different cleaner and not directly comparable.
• When enabling both bf16 and torch.compile, the first inference may be very slow (90s) due to compilation overhead. Subsequent inferences will be significantly faster (58s) as the compiled artifacts are cached.