search

PaddlePaddle / PaddleDetection

Object Detection toolkit based on PaddlePaddle. It supports object detection, instance segmentation, multiple object tracking and real-time multi-person keypoint detection.
Apache License 2.0
12.83k stars 2.89k forks source link

RuntimeError: (NotFound) Cannot open file C:\Users\Abc/.cache/paddle/infer_weights\STGCN, please confirm whether the file is normal.[Hint: Expected static_cast<bool>(fin.is_open()) == true, but received static_cast<bool>(fin.is_open()):0 != true:1.] (at ..\paddle\fluid\inference\api\analysis_predictor.cc:2808) #9202

Open DazzlingGalaxy opened 2 weeks ago

DazzlingGalaxy commented 2 weeks ago

问题确认 Search before asking

Bug组件 Bug Component

Inference

Bug描述 Describe the Bug

新装的PaddleDetection环境，下载仓库develop分支zip代码后安装的。想试一下打架识别，本地已准备了一个mp4文件。参考文档中的用法，运行后报错 python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_pphuman.yml --video_file=z:/cam1_9.mp4 --device=gpu


  File "deploy/pipeline/pipeline.py", line 1321, in <module>
    main()
  File "deploy/pipeline/pipeline.py", line 1306, in main
    pipeline = Pipeline(FLAGS, cfg)
  File "deploy/pipeline/pipeline.py", line 97, in __init__
    self.predictor = PipePredictor(args, cfg, self.is_video)
  File "deploy/pipeline/pipeline.py", line 439, in __init__
    self.skeleton_action_predictor = SkeletonActionRecognizer.init_with_cfg(
  File "Z:\PaddleDetection-develop\deploy\pipeline\pphuman\action_infer.py", line 92, in init_with_cfg
    return cls(model_dir=cfg['model_dir'],
  File "Z:\PaddleDetection-develop\deploy\pipeline\pphuman\action_infer.py", line 75, in __init__
    super(SkeletonActionRecognizer, self).__init__(
  File "Z:\PaddleDetection-develop\deploy\python\infer.py", line 108, in __init__
    self.predictor, self.config = load_predictor(
  File "Z:\PaddleDetection-develop\deploy\python\infer.py", line 1098, in load_predictor
    predictor = create_predictor(config)
RuntimeError: (NotFound) Cannot open file C:\Users\Abc/.cache/paddle/infer_weights\STGCN, please confirm whether the file is normal.
  [Hint: Expected static_cast<bool>(fin.is_open()) == true, but received static_cast<bool>(fin.is_open()):0 != true:1.] (at ..\paddle\fluid\inference\api\analysis_predictor.cc:2808)```

`STGCN`文件夹存在且里面有内容
![1](https://github.com/user-attachments/assets/962ef414-0f1d-4298-94e3-d3510da9d898)

这个问题之前有人提过，但没解决， #8287 

顺便再问一个以前有人问但没解决的问题，里面有人说可以手动下载文件，但不知道去哪下载，也不知道在哪个文件能看到它需要的下载地址， #9192 

### 复现环境 Environment

win11

anyio                    4.5.2
astor                    0.8.1
autocommand              2.2.2
babel                    2.16.0
backports.tarfile        1.2.0
bce-python-sdk           0.9.23
blinker                  1.8.2
certifi                  2024.8.30
charset-normalizer       3.4.0
click                    8.1.7
colorama                 0.4.6
contourpy                1.1.1
cycler                   0.12.1
Cython                   3.0.11
decorator                5.1.1
exceptiongroup           1.2.2
Flask                    3.0.3
flask-babel              4.0.0
fonttools                4.54.1
future                   1.0.0
h11                      0.14.0
httpcore                 1.0.6
httpx                    0.27.2
idna                     3.10
imageio                  2.35.1
imgaug                   0.4.0
importlib_metadata       8.5.0
importlib_resources      6.4.5
inflect                  7.3.1
itsdangerous             2.2.0
jaraco.collections       5.1.0
jaraco.context           5.3.0
jaraco.functools         4.0.1
jaraco.text              3.12.1
Jinja2                   3.1.4
joblib                   1.4.2
kiwisolver               1.4.7
lapx                     0.5.11
lazy_loader              0.4
MarkupSafe               2.1.5
matplotlib               3.7.5
more-itertools           10.3.0
motmetrics               1.4.0
networkx                 3.1
numpy                    1.24.4
nvidia-cublas-cu11       11.11.3.6
nvidia-cuda-nvrtc-cu11   11.8.89
nvidia-cuda-runtime-cu11 11.8.89
nvidia-cudnn-cu11        8.9.4.19
nvidia-cufft-cu11        10.9.0.58
nvidia-curand-cu11       10.3.0.86
nvidia-cusolver-cu11     11.4.1.48
nvidia-cusparse-cu11     11.7.5.86
opencv-python            4.5.5.64
opt-einsum               3.3.0
packaging                24.1
paddledet                0.0.0
paddlepaddle-gpu         3.0.0b1
pandas                   2.0.3
pillow                   10.4.0
pip                      24.2
platformdirs             4.2.2
protobuf                 5.28.3
psutil                   6.1.0
pyclipper                1.3.0.post6
pycocotools              2.0.7
pycryptodome             3.21.0
pyparsing                3.1.4
python-dateutil          2.9.0.post0
pytz                     2024.2
PyWavelets               1.4.1
PyYAML                   6.0.2
rarfile                  4.2
requests                 2.32.3
scikit-image             0.21.0
scikit-learn             1.3.2
scipy                    1.10.1
setuptools               75.1.0
shapely                  2.0.6
six                      1.16.0
sklearn                  0.0
sniffio                  1.3.1
terminaltables           3.1.10
threadpoolctl            3.5.0
tifffile                 2023.7.10
tomli                    2.0.1
tqdm                     4.66.6
typeguard                4.4.0
typing_extensions        4.12.2
tzdata                   2024.2
urllib3                  2.2.3
visualdl                 2.5.3
Werkzeug                 3.0.6
wheel                    0.44.0
xmltodict                0.14.2
zipp                     3.20.2

### Bug描述确认 Bug description confirmation

- [X] 我确认已经提供了Bug复现步骤、代码改动说明、以及环境信息，确认问题是可以复现的。I confirm that the bug replication steps, code change instructions, and environment information have been provided, and the problem can be reproduced.

### 是否愿意提交PR？ Are you willing to submit a PR?

- [ ] 我愿意提交PR！I'd like to help by submitting a PR!
DazzlingGalaxy commented 2 weeks ago

@TingquanGao ，更新一下，之前报错的文件夹是paddle/infer_weights/STGCN，我发现它其实是文档中的摔倒识别。摔倒识别和打架识别昨天我都试了，可能是先试的摔倒识别，设置infer_cfg_pphuman.yml中SKELETON_ACTION的enable: True，然后发现报错，没把它再设为False就接着把打架识别的VIDEO_ACTION设为True了，所以昨天我说想用打架识别但报错STGCN文件夹。

今天我做了下面的操作： 1.基于骨骼点的行为识别之摔倒识别，只开启SKELETON_ACTION，其他都是False，报错； 2.基于图像分类的行为识别之行人检测，只开启ID_BASED_CLSACTION，其他都是False，报错； 3.基于检测的行为识别之行人检测，只开启ID_BASED_DETACTION，其他都是False，报错； 4.基于视频分类的行为识别之打架识别，只开启VIDEO_ACTION，其他都是False，正常，输入一个视频文件，会输出一个视频文件，并在打架的时刻标出置信度。 前三次的报错都和昨天报错STGCN文件夹类似，都是paddle/infer_weights/xxx这种。

我想问几点： 1.前三种报错，如何解决？ 2.打架识别，只能输入一个视频文件吗？我试了一张图片会报错

  File "deploy/pipeline/pipeline.py", line 1321, in <module>
    main()
  File "deploy/pipeline/pipeline.py", line 1308, in main
    pipeline.run_multithreads()
  File "deploy/pipeline/pipeline.py", line 179, in run_multithreads
    self.predictor.run(self.input)
  File "deploy/pipeline/pipeline.py", line 533, in run
    self.predict_video(input, thread_idx=thread_idx)
  File "deploy/pipeline/pipeline.py", line 986, in predict_video
    if frame_id % sample_freq == 0:
ZeroDivisionError: integer division or modulo by zero

3.打架识别完成后，除了手动打开生成的视频文件，怎么知道视频中有没有打架？ 4.我的实际需求是，我有一些摄像头，要么我直接从摄像头获取流数据或画面(图片)，要么其他人去摄像头获取，我再从他那去获取，然后我识别画面中有没有打架。PaddleDetection能直接识别摄像头数据吗(不管是我直接从摄像头获取，还是我从其他人那获取)？能的话有没有例子？不能的话，我应该频繁将摄像头画面保存成图片，然后合成一小段视频(比如5分钟的视频)，再传给PaddleDetection？这样时效性感觉有点滞后。还是说有其他更好的办法？

TingquanGao commented 2 weeks ago
  1. 可以尝试手动下载模型(文档)并修改配置文件，可以参考文档；
  2. 行为识别模块仅支持视频输入；
  3. 如果要保存下来，需要改下代码；
  4. 请参考文档。
DazzlingGalaxy commented 2 weeks ago
  1. 可以尝试手动下载模型(文档)并修改配置文件，可以参考文档；
  2. 行为识别模块仅支持视频输入；
  3. 如果要保存下来，需要改下代码；
  4. 请参考文档。

针对1.模型文件其实是在的，文件夹中有文件(昨天也发了图片，但没显示出来，只显示了链接)，是自动下载的，但还是报错，比如这些 1 2

针对3①.怎么修改代码？目前我是用subprocess.run()去运行deploy/pipeline/pipeline.py，然后获取打印内容，再判断打印内容中有没有video_action_res: {'class': 1, 'score': 0.7190255}，有的话说明有打架。 针对3②.运行deploy/pipeline/pipeline.py后有一些打印，意思是frame id在40-50和100-110之间时有打架吗？

video fps: 30, frame_count: 160
Thread: 0; frame id: 0
Thread: 0; frame id: 10
Thread: 0; frame id: 20
Thread: 0; frame id: 30
Thread: 0; frame id: 40
W1105 09:04:42.407253 12872 gpu_resources.cc:119] Please NOTE: device: 0, GPU Compute Capability: 8.6, Driver API Version: 12.3, Runtime API Version: 11.8
W1105 09:04:42.407253 12872 gpu_resources.cc:164] device: 0, cuDNN Version: 8.9.
I1105 09:04:42.407253 12872 program_interpreter.cc:243] New Executor is Running.
video_action_res: {'class': 1, 'score': 0.5839483}
Thread: 0; frame id: 50
Thread: 0; frame id: 60
Thread: 0; frame id: 70
Thread: 0; frame id: 80
Thread: 0; frame id: 90
Thread: 0; frame id: 100
video_action_res: {'class': 1, 'score': 0.7190255}
Thread: 0; frame id: 110
Thread: 0; frame id: 120
Thread: 0; frame id: 130
Thread: 0; frame id: 140
Thread: 0; frame id: 150
save result to output\cam1_9.mp4
------------------ Inference Time Info ----------------------
total_time(ms): 150.2, img_num: 109
video_action time(ms): 60.1; per frame average time(ms): 0.5513761467889908
average latency time(ms): 1.38, QPS: 725.699068

针对3③.python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_pphuman.yml --video_file=z:/cam1_9.mp4 --device=gpu，完整的打印信息如下，我运行的正常吗？比如它默认申请8G显存，我只有6G，会有影响吗？可以手动设置申请的显存大小吗？

-----------  Running Arguments -----------
ATTR:
  batch_size: 8
  enable: false
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/PPLCNet_x1_0_person_attribute_945_infer.zip
DET:
  batch_size: 1
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip
ID_BASED_CLSACTION:
  batch_size: 8
  display_frames: 80
  enable: false
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/PPHGNet_tiny_calling_halfbody.zip
  skip_frame_num: 2
  threshold: 0.8
ID_BASED_DETACTION:
  batch_size: 8
  display_frames: 80
  enable: false
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/ppyoloe_crn_s_80e_smoking_visdrone.zip
  skip_frame_num: 2
  threshold: 0.6
KPT:
  batch_size: 8
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip
MOT:
  batch_size: 1
  enable: false
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip
  skip_frame_num: -1
  tracker_config: deploy/pipeline/config/tracker_config.yml
REID:
  batch_size: 16
  enable: false
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/reid_model.zip
SKELETON_ACTION:
  batch_size: 1
  coord_size:
  - 384
  - 512
  display_frames: 80
  enable: false
  max_frames: 50
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/STGCN.zip
VIDEO_ACTION:
  batch_size: 1
  enable: true
  frame_len: 8
  model_dir: https://videotag.bj.bcebos.com/PaddleVideo-release2.3/ppTSM_fight.zip
  sample_freq: 7
  short_size: 340
  target_size: 320
attr_thresh: 0.5
crop_thresh: 0.5
kpt_thresh: 0.2
visual: true
warmup_frame: 50

------------------------------------------
VideoAction Recognition enabled
DET  model dir:  C:\Users\Abc/.cache/paddle/infer_weights\mot_ppyoloe_l_36e_pipeline
mot_model_dir model_dir:  C:\Users\Abc/.cache/paddle/infer_weights\mot_ppyoloe_l_36e_pipeline
KPT  model dir:  C:\Users\Abc/.cache/paddle/infer_weights\dark_hrnet_w32_256x192
VIDEO_ACTION  model dir:  C:\Users\Abc/.cache/paddle/infer_weights\ppTSM
E1105 11:49:36.710119 23412 analysis_predictor.cc:2137] Allocate too much memory for the GPU memory pool, assigned 8000 MB
E1105 11:49:36.710119 23412 analysis_predictor.cc:2140] Try to shrink the value by setting AnalysisConfig::EnableUseGpu(...)
e[1me[35m--- Running analysis [ir_graph_build_pass]e[0m
I1105 11:49:36.738330 23412 executor.cc:184] Old Executor is Running.
e[1me[35m--- Running analysis [ir_analysis_pass]e[0m
e[32m--- Running IR pass [map_op_to_another_pass]e[0m
I1105 11:49:36.824437 23412 fuse_pass_base.cc:59] ---  detected 55 subgraphs
e[32m--- Running IR pass [is_test_pass]e[0m
e[32m--- Running IR pass [simplify_with_basic_ops_pass]e[0m
e[32m--- Running IR pass [delete_quant_dequant_linear_op_pass]e[0m
e[32m--- Running IR pass [delete_weight_dequant_linear_op_pass]e[0m
e[32m--- Running IR pass [constant_folding_pass]e[0m
e[32m--- Running IR pass [silu_fuse_pass]e[0m
e[32m--- Running IR pass [conv_bn_fuse_pass]e[0m
I1105 11:49:36.929447 23412 fuse_pass_base.cc:59] ---  detected 55 subgraphs
e[32m--- Running IR pass [conv_eltwiseadd_bn_fuse_pass]e[0m
e[32m--- Running IR pass [embedding_eltwise_layernorm_fuse_pass]e[0m
e[32m--- Running IR pass [multihead_matmul_fuse_pass_v2]e[0m
e[32m--- Running IR pass [vit_attention_fuse_pass]e[0m
e[32m--- Running IR pass [fused_multi_transformer_encoder_pass]e[0m
e[32m--- Running IR pass [fused_multi_transformer_decoder_pass]e[0m
e[32m--- Running IR pass [fused_multi_transformer_encoder_fuse_qkv_pass]e[0m
e[32m--- Running IR pass [fused_multi_transformer_decoder_fuse_qkv_pass]e[0m
e[32m--- Running IR pass [multi_devices_fused_multi_transformer_encoder_pass]e[0m
e[32m--- Running IR pass [multi_devices_fused_multi_transformer_encoder_fuse_qkv_pass]e[0m
e[32m--- Running IR pass [multi_devices_fused_multi_transformer_decoder_fuse_qkv_pass]e[0m
e[32m--- Running IR pass [fuse_multi_transformer_layer_pass]e[0m
e[32m--- Running IR pass [gpu_cpu_squeeze2_matmul_fuse_pass]e[0m
e[32m--- Running IR pass [gpu_cpu_reshape2_matmul_fuse_pass]e[0m
e[32m--- Running IR pass [gpu_cpu_flatten2_matmul_fuse_pass]e[0m
e[32m--- Running IR pass [gpu_cpu_map_matmul_v2_to_mul_pass]e[0m
I1105 11:49:38.095309 23412 fuse_pass_base.cc:59] ---  detected 1 subgraphs
e[32m--- Running IR pass [gpu_cpu_map_matmul_v2_to_matmul_pass]e[0m
e[32m--- Running IR pass [matmul_scale_fuse_pass]e[0m
e[32m--- Running IR pass [multihead_matmul_fuse_pass_v3]e[0m
e[32m--- Running IR pass [gpu_cpu_map_matmul_to_mul_pass]e[0m
e[32m--- Running IR pass [fc_fuse_pass]e[0m
I1105 11:49:38.129885 23412 fuse_pass_base.cc:59] ---  detected 1 subgraphs
e[32m--- Running IR pass [fc_elementwise_layernorm_fuse_pass]e[0m
e[32m--- Running IR pass [conv_elementwise_add_act_fuse_pass]e[0m
e[32m--- Running IR pass [conv_elementwise_add2_act_fuse_pass]e[0m
e[32m--- Running IR pass [conv_elementwise_add_fuse_pass]e[0m
I1105 11:49:38.204499 23412 fuse_pass_base.cc:59] ---  detected 55 subgraphs
e[32m--- Running IR pass [transpose_flatten_concat_fuse_pass]e[0m
e[32m--- Running IR pass [transfer_layout_pass]e[0m
e[32m--- Running IR pass [transfer_layout_elim_pass]e[0m
e[32m--- Running IR pass [auto_mixed_precision_pass]e[0m
e[32m--- Running IR pass [identity_op_clean_pass]e[0m
e[32m--- Running IR pass [inplace_op_var_pass]e[0m
I1105 11:49:38.212951 23412 fuse_pass_base.cc:59] ---  detected 2 subgraphs
e[1me[35m--- Running analysis [ir_params_sync_among_devices_pass]e[0m
I1105 11:49:38.214633 23412 ir_params_sync_among_devices_pass.cc:51] Sync params from CPU to GPU
e[1me[35m--- Running analysis [adjust_cudnn_workspace_size_pass]e[0m
e[1me[35m--- Running analysis [inference_op_replace_pass]e[0m
e[1me[35m--- Running analysis [save_optimized_model_pass]e[0m
e[1me[35m--- Running analysis [ir_graph_to_program_pass]e[0m
I1105 11:49:38.334097 23412 analysis_predictor.cc:2080] ======= ir optimization completed =======
I1105 11:49:38.334097 23412 naive_executor.cc:200] ---  skip [feed], feed -> data_batch_0
I1105 11:49:38.335599 23412 naive_executor.cc:200] ---  skip [linear_2.tmp_1], fetch -> fetch
video fps: 30, frame_count: 160
Thread: 0; frame id: 0
Thread: 0; frame id: 10
Thread: 0; frame id: 20
Thread: 0; frame id: 30
Thread: 0; frame id: 40
W1105 11:49:40.078011 23412 gpu_resources.cc:119] Please NOTE: device: 0, GPU Compute Capability: 8.6, Driver API Version: 12.3, Runtime API Version: 11.8
W1105 11:49:40.078011 23412 gpu_resources.cc:164] device: 0, cuDNN Version: 8.9.
I1105 11:49:40.078011 23412 program_interpreter.cc:243] New Executor is Running.
video_action_res: {'class': 1, 'score': 0.5839483}
Thread: 0; frame id: 50
Thread: 0; frame id: 60
Thread: 0; frame id: 70
Thread: 0; frame id: 80
Thread: 0; frame id: 90
Thread: 0; frame id: 100
video_action_res: {'class': 1, 'score': 0.7190255}
Thread: 0; frame id: 110
Thread: 0; frame id: 120
Thread: 0; frame id: 130
Thread: 0; frame id: 140
Thread: 0; frame id: 150
save result to output\cam1_9.mp4
------------------ Inference Time Info ----------------------
total_time(ms): 216.6, img_num: 109
video_action time(ms): 80.10000000000001; per frame average time(ms): 0.734862385321101
average latency time(ms): 1.99, QPS: 503.231764

针对4.之前看到rtsp的方式了，不过发issue时忘了这点，刚试了下报错(单独用opencv连接rtsp，可以获取到摄像头画面)，我没修改deploy/pipeline/config/examples/infer_cfg_human_attr.yml中的内容，python deploy/pipeline/pipeline.py --config deploy/pipeline/config/examples/infer_cfg_human_attr.yml -o visual=False --rtsp rtsp://abc:abc@192.168.0.1:554/Streaming/Channels/101 --device=gpu

Traceback (most recent call last):
  File "deploy/pipeline/pipeline.py", line 1321, in <module>
    main()
  File "deploy/pipeline/pipeline.py", line 1303, in main
    cfg = merge_cfg(FLAGS)  # use command params to update config
  File "Z:\PaddleDetection-develop\deploy\pipeline\cfg_utils.py", line 212, in merge_cfg
    pred_config = merge_opt(pred_config, args_dict)
  File "Z:\PaddleDetection-develop\deploy\pipeline\cfg_utils.py", line 202, in merge_opt
    for sub_k, sub_v in value.items():
AttributeError: 'bool' object has no attribute 'items'

还试了python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_pphuman.yml -o visual=False --rtsp rtsp://abc:abc@192.168.0.1:554/Streaming/Channels/101 --device=gpu，也报错，deploy/pipeline/config/infer_cfg_pphuman.yml中只有VIDEO_ACTION是True，其他是False

Traceback (most recent call last):
  File "deploy/pipeline/pipeline.py", line 1321, in <module>
    main()
  File "deploy/pipeline/pipeline.py", line 1303, in main
    cfg = merge_cfg(FLAGS)  # use command params to update config
  File "Z:\PaddleDetection-develop\deploy\pipeline\cfg_utils.py", line 212, in merge_cfg
    pred_config = merge_opt(pred_config, args_dict)
  File "Z:\PaddleDetection-develop\deploy\pipeline\cfg_utils.py", line 202, in merge_opt
    for sub_k, sub_v in value.items():
AttributeError: 'bool' object has no attribute 'items'

还试了python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_pphuman.yml --video_file=rtsp://abc:abc@192.168.0.1:554/Streaming/Channels/101 --device=gpu，因为文档中说或者video_file后面的视频地址直接更换为rtsp流地址，不报错，但是为什么运行后只处理frame id为0-30的数据？处理完后就没了，程序也没停止运行

video fps: 20, frame_count: -2049638230412172
Thread: 0; frame id: 0
Thread: 0; frame id: 10
Thread: 0; frame id: 20
Thread: 0; frame id: 30
save result to output\101_t00_rtsp.mp4
------------------ Inference Time Info ----------------------
total_time(ms): 0.0, img_num: 0
average latency time(ms): 0.00, QPS: 0.000000
DazzlingGalaxy commented 2 weeks ago

@TingquanGao 可以解答下我上面又发的吗？

TingquanGao commented 1 week ago
  1. 参考文档修改配置文件，传入模型本地路径，是否可以？
  2. 保存的逻辑具体可以看下代码，看起来是在这里；
  3. 打印的video_action_res: {'class': 1, 'score': 0.7190255}不是说明在这之间检测到了打架行为，打架行为是使用了最近50帧的视频进行识别的，如果识别到了就会打印，可以看下相关代码；
  4. 比如它默认申请8G显存，我只有6G：你提供的配置中，我没看到有8G显存的设置；
  5. 视频流的问题我得再问下其他人。
DazzlingGalaxy commented 5 days ago
  1. 参考文档修改配置文件，传入模型本地路径，是否可以？
  2. 保存的逻辑具体可以看下代码，看起来是在这里；
  3. 打印的video_action_res: {'class': 1, 'score': 0.7190255}不是说明在这之间检测到了打架行为，打架行为是使用了最近50帧的视频进行识别的，如果识别到了就会打印，可以看下相关代码；
  4. 比如它默认申请8G显存，我只有6G：你提供的配置中，我没看到有8G显存的设置；
  5. 视频流的问题我得再问下其他人。

最近一直在外地，我有时间了再试下，感谢。8G显存是我前面贴出的打印，Allocate too much memory for the GPU memory pool, assigned 8000 MB