Overlap detect and clean

[linkedlist771]

We can overlap detect and remove once the cuda mfu is lower.

[linkedlist771]

From branch, we can see it did provide about 16% increase. It seems work, after os check, we could merge it into the main branch.

Analysis:
This overlap increase is indeed valid because most of the time consuming operations are about the CPU preprocess in detect, it can be overlapped with the GPU ops.

2025-10-13 21:07:47.267 | DEBUG    | sorawm.pipeline:worker_main:46 - [DetectorWorker] Starting detector process
2025-10-13 21:07:47.267 | DEBUG    | sorawm.watermark_detector:__init__:18 - Begin to load yolo water mark detet model.
2025-10-13 21:07:47.273 | DEBUG    | sorawm.pipeline:worker_main:97 - [CleanerWorker] Starting cleaner process
2025-10-13 21:07:47.273 | DEBUG    | sorawm.utils.devices_utils:get_device:13 - Using device: cuda
2025-10-13 21:07:47.275 | INFO     | sorawm.iopaint.model_manager:init_model:47 - Loading model: lama
2025-10-13 21:07:47.291 | DEBUG    | sorawm.utils.devices_utils:get_device:13 - Using device: cuda
2025-10-13 21:07:47.378 | DEBUG    | sorawm.watermark_detector:__init__:21 - Yolo water mark detet model loaded.
Processing frames:   8%|█████▏                                                               | 21/280 [00:06<00:36,  7.17it/s]2025-10-13 21:07:53.388 | DEBUG    | sorawm.pipeline:worker_main:130 - [CleanerWorker] Frame 22: Using last valid bbox
Processing frames:  52%|███████████████████████████████████▋                                | 147/280 [00:21<00:14,  8.92it/s]2025-10-13 21:08:07.335 | DEBUG    | sorawm.pipeline:worker_main:130 - [CleanerWorker] Frame 148: Using last valid bbox
Processing frames:  78%|█████████████████████████████████████████████████████▏              | 219/280 [00:29<00:06,  9.23it/s]2025-10-13 21:08:15.215 | INFO     | sorawm.core:input_worker:201 - [Pipeline] Input worker: All frames fed
Processing frames:  89%|████████████████████████████████████████████████████████████▏       | 248/280 [00:33<00:03,  8.88it/s]2025-10-13 21:08:18.542 | DEBUG    | sorawm.pipeline:worker_main:58 - [DetectorWorker] Received shutdown signal
Processing frames:  89%|████████████████████████████████████████████████████████████▍       | 249/280 [00:33<00:03,  8.89it/s]2025-10-13 21:08:18.652 | DEBUG    | sorawm.pipeline:worker_main:81 - [DetectorWorker] Exiting
Processing frames:  98%|██████████████████████████████████████████████████████████████████▎ | 273/280 [00:35<00:00,  9.01it/s]2025-10-13 21:08:21.316 | DEBUG    | sorawm.pipeline:worker_main:130 - [CleanerWorker] Frame 274: Using last valid bbox
Processing frames: 100%|███████████████████████████████████████████████████████████████████▊| 279/280 [00:36<00:00,  9.07it/s2025-10-13 21:08:21.973 | DEBUG    | sorawm.pipeline:worker_main:115 - [CleanerWorker] Received shutdown signal, processed 280 frames
2025-10-13 21:08:21.973 | DEBUG    | sorawm.pipeline:worker_main:174 - [CleanerWorker] Exiting
Processing frames: 100%|████████████████████████████████████████████████████████████████████| 280/280 [00:36<00:00,  7.63it/s]
2025-10-13 21:08:21.980 | INFO     | sorawm.core:output_worker:232 - [Pipeline] Output worker: Processed 280 frames total
2025-10-13 21:08:22.219 | INFO     | sorawm.core:merge_audio_track:273 - Merging audio track...
2025-10-13 21:08:22.319 | INFO     | sorawm.core:merge_audio_track:290 - Saved no watermark video with audio at: outputs/benchmark/benchmark_parallel.mp4
2025-10-13 21:08:22.319 | INFO     | sorawm.pipeline:shutdown:235 - [PipelineManager] Shutting down pipeline...
2025-10-13 21:08:22.444 | INFO     | sorawm.pipeline:shutdown:250 - [PipelineManager] Pipeline shutdown complete
2025-10-13 21:08:22.444 | INFO     | __main__:benchmark_method:43 - 
run_overlap completed!
2025-10-13 21:08:22.444 | INFO     | __main__:benchmark_method:44 - Time elapsed: 37.18 seconds
2025-10-13 21:08:22.444 | INFO     | __main__:benchmark_method:45 - Output file: outputs/benchmark/benchmark_parallel.mp4
2025-10-13 21:08:22.444 | INFO     | __main__:main:103 - 
============================================================
2025-10-13 21:08:22.444 | INFO     | __main__:main:104 - Performance Comparison Summary
2025-10-13 21:08:22.444 | INFO     | __main__:main:105 - ============================================================
2025-10-13 21:08:22.444 | INFO     | __main__:main:108 - Serial execution (run):           44.60 seconds
2025-10-13 21:08:22.444 | INFO     | __main__:main:109 - Pipeline parallel (run_overlap):  37.18 seconds
2025-10-13 21:08:22.444 | INFO     | __main__:main:110 - 
2025-10-13 21:08:22.444 | INFO     | __main__:main:117 - ✅ Performance improvement: 1.20x speedup
2025-10-13 21:08:22.444 | INFO     | __main__:main:118 - ✅ Time reduction: 16.7%
2025-10-13 21:08:22.444 | INFO     | __main__:main:119 - ✅ Time saved: 7.43 seconds
2025-10-13 21:08:22.444 | INFO     | __main__:main:128 - 
============================================================
2025-10-13 21:08:22.444 | INFO     | __main__:main:129 - Benchmark Complete
2025-10-13 21:08:22.444 | INFO     | __main__:main:130 - ============================================================

[kcxain]

Awesome!

[linkedlist771]

With the missing detect imputation feature. The overlap becomes somehow redundant sometimes, this feature is set as draft now.