记录下在windows中使用mmrotate1.x中的benchmark测试fps踩过的坑

Pliter

已于 2024-09-04 00:53:48 修改

阅读量571

点赞数 3

CC 4.0 BY-SA版权

文章标签：人工智能深度学习

于 2024-08-31 16:36:43 首次发布

本文链接：https://blog.youkuaiyun.com/Pliter/article/details/141754441

Windows下使用mmrotate1.x中tools/analysis_tools/benchmark.py测试fps报错如下：

Traceback (most recent call last):
  File "tools/analysis_tools/benchmark.py", line 133, in <module>
    main()
  File "tools/analysis_tools/benchmark.py", line 128, in main
    benchmark = eval(f'{args.task}_benchmark')(args, cfg, distributed, logger)
  File "tools/analysis_tools/benchmark.py", line 72, in inference_benchmark
    benchmark = InferenceBenchmark(
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py", line 161, in __init__
    print_process_memory(self._process, self.logger)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py", line 53, in print_process_memory
    pss_mem = gb_round(memory_full_info.pss)
AttributeError: 'pfullmem' object has no attribute 'pss'

进入"H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py"文件，将报错处代码的pss改为rss，具体如下

def print_process_memory(p: psutil.Process,
                         logger: Optional[MMLogger] = None) -> None:
    """print process memory info."""
    mem_used = gb_round(psutil.virtual_memory().used)
    memory_full_info = p.memory_full_info()
    uss_mem = gb_round(memory_full_info.uss)

    # pss_mem = gb_round(memory_full_info.pss)
    pss_mem = gb_round(memory_full_info.rss)        #############
    for children in p.children():
        child_mem_info = children.memory_full_info()
        uss_mem += gb_round(child_mem_info.uss)

        # pss_mem += gb_round(child_mem_info.pss)
        pss_mem += gb_round(child_mem_info.rss)     #############共修改两处
    process_count = 1 + len(p.children())
    print_log(
        f'(GB) mem_used: {mem_used:.2f} | uss: {uss_mem:.2f} | '
        f'pss: {pss_mem:.2f} | total_proc: {process_count}', logger)

改完后又会报错：

Traceback (most recent call last):
  File "tools/analysis_tools/benchmark.py", line 134, in <module>
    main()
  File "tools/analysis_tools/benchmark.py", line 129, in main
    benchmark = eval(f'{args.task}_benchmark')(args, cfg, distributed, logger)
  File "tools/analysis_tools/benchmark.py", line 72, in inference_benchmark
    benchmark = InferenceBenchmark(
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py", line 164, in __init__
    self.model = self._init_model(checkpoint, is_fuse_conv_bn)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py", line 180, in _init_model
    model = MODELS.build(self.cfg.model)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmengine\registry\registry.py", line 570, in build
    return self.build_func(cfg, *args, **kwargs, registry=self)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmengine\registry\build_functions.py", line 232, in build_model_from_cfg
    return build_from_cfg(cfg, registry, default_args)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmengine\registry\build_functions.py", line 121, in build_from_cfg
    obj = obj_cls(**args)  # type: ignore
TypeError: __init__() got an unexpected keyword argument 'pretrained'

还是在上面那个文件里，将InferenceBenchmark(BaseBenchmark)类中的这段代码删掉或注释掉。

self.cfg.model.pretrained = None

接着报错：

Traceback (most recent call last):
  File "tools/analysis_tools/benchmark.py", line 133, in <module>
    main()
  File "tools/analysis_tools/benchmark.py", line 129, in main
    benchmark.run(args.repeat_num)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py", line 100, in run
    results.append(self.run_once())
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\mmdet\utils\benchmark.py", line 216, in run_once
    self.model(data, return_loss=False)
  File "H:\Anaconda\envs\mmrotate1.x\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
TypeError: forward() got an unexpected keyword argument 'return_loss'

还是在这个类中，改动def run_once(self)函数中的这段代码

            with torch.no_grad():
                # self.model(data, return_loss=False)   #######把这句改成下句
                self.model.test_step(data)

最后运行指令

python tools/analysis_tools/benchmark.py work_dirs/test/config.py --checkpoint work_dirs/test/epoch_120.pth

(mmrotate1.x) PS H:\mmrotate1.x> python tools/analysis_tools/benchmark.py work_dirs/test/config.py --checkpoint work_dirs/test/epoch_120.pth
08/31 16:13:08 - mmengine - INFO - before build: 
08/31 16:13:08 - mmengine - INFO - (GB) mem_used: 10.56 | uss: 0.15 | pss: 0.23 | total_proc: 1
Loads checkpoint by local backend from path: work_dirs/test/epoch_120.pth
08/31 16:13:09 - mmengine - INFO - after build: 
08/31 16:13:09 - mmengine - INFO - (GB) mem_used: 11.98 | uss: 1.44 | pss: 1.63 | total_proc: 1
08/31 16:13:13 - mmengine - INFO - ==================================
08/31 16:13:13 - mmengine - INFO - Done image [50 /2000], fps: 69.7 img/s, times per image: 14.3 ms/img, cuda memory: 51 MB
08/31 16:13:13 - mmengine - INFO - (GB) mem_used: 13.51 | uss: 2.78 | pss: 3.27 | total_proc: 1
08/31 16:13:15 - mmengine - INFO - ==================================
08/31 16:13:15 - mmengine - INFO - Done image [100/2000], fps: 69.5 img/s, times per image: 14.4 ms/img, cuda memory: 51 MB
08/31 16:13:15 - mmengine - INFO - (GB) mem_used: 13.53 | uss: 2.78 | pss: 3.27 | total_proc: 1
08/31 16:13:16 - mmengine - INFO - ==================================
08/31 16:13:16 - mmengine - INFO - Done image [150/2000], fps: 67.7 img/s, times per image: 14.8 ms/img, cuda memory: 51 MB
08/31 16:13:16 - mmengine - INFO - (GB) mem_used: 13.53 | uss: 2.78 | pss: 3.27 | total_proc: 1
08/31 16:13:17 - mmengine - INFO - ==================================
08/31 16:13:17 - mmengine - INFO - Done image [200/2000], fps: 65.2 img/s, times per image: 15.3 ms/img, cuda memory: 51 MB
08/31 16:13:17 - mmengine - INFO - (GB) mem_used: 13.51 | uss: 2.78 | pss: 3.27 | total_proc: 1
08/31 16:13:18 - mmengine - INFO - ============== Done ==================
08/31 16:13:18 - mmengine - INFO - Overall fps: 65.2 img/s, times per image: 15.3 ms/img
08/31 16:13:18 - mmengine - INFO - cuda memory: 51 MB
08/31 16:13:18 - mmengine - INFO - (GB) mem_used: 13.52 | uss: 2.78 | pss: 3.26 | total_proc: 1

成功计算出fps