PP-Human v2 自定义目标检测模型不读取类别

[KirigiriSuzumiya]

问题确认 Search before asking

• [X] 我已经查询历史issue，没有报过同样bug。I have searched the issues and found no similar bug report.

bug描述 Describe the Bug

• 替换了mot模型地址为自己的ppyoloe模型，可以正常预测但是可视化后的视频检测框中仍然显示pedestrian。

• 另外，在替换配置文件infer_cfg_pphuman.yml中的ID_BASED_DETACTION模型地址后，报错：(这个报错只出现在我自定义的模型检测到目标时，前置的mot任务似乎正常)

  Traceback (most recent call last):
  File "C:\Users\boyif\Desktop\paddle\PaddleDetection\deploy\pipeline\pipeline.py", line 1098, in <module>
  main()
  File "C:\Users\boyif\Desktop\paddle\PaddleDetection\deploy\pipeline\pipeline.py", line 1087, in main
  pipeline.run()
  File "C:\Users\boyif\Desktop\paddle\PaddleDetection\deploy\pipeline\pipeline.py", line 175, in run
  self.predictor.run(self.input)
  File "C:\Users\boyif\Desktop\paddle\PaddleDetection\deploy\pipeline\pipeline.py", line 585, in run
  self.predict_video(input)
  File "C:\Users\boyif\Desktop\paddle\PaddleDetection\deploy\pipeline\pipeline.py", line 801, in predict_video
  self.pipeline_res.update(det_action_res, 'det_action')
  File "C:\Users\boyif\Desktop\paddle\PaddleDetection\deploy\pipeline\datacollector.py", line 36, in update
  self.res_dict[name].update(res)
  ValueError: dictionary update sequence element #0 has length 0; 2 is required

怀疑是没有读取目标检测模型的类别标签导致的两个问题。

复现环境 Environment

infer_cfg_pphuman.yml

crop_thresh: 0.5
attr_thresh: 0.5
kpt_thresh: 0.2
visual: True
warmup_frame: 50

DET:
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip
  batch_size: 1

MOT:
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/mot_ppyoloe_l_36e_pipeline.zip
  tracker_config: deploy/pipeline/config/tracker_config.yml
  batch_size: 1
  basemode: "idbased"
  enable: True

KPT:
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/dark_hrnet_w32_256x192.zip
  batch_size: 8

ATTR:
  model_dir:  https://bj.bcebos.com/v1/paddledet/models/pipeline/PPLCNet_x1_0_person_attribute_945_infer.zip
  batch_size: 8
  basemode: "idbased"
  enable: False

VIDEO_ACTION:
  model_dir: https://videotag.bj.bcebos.com/PaddleVideo-release2.3/ppTSM_fight.zip
  batch_size: 1
  frame_len: 8
  sample_freq: 7
  short_size: 340
  target_size: 320
  basemode: "videobased"
  enable: False

SKELETON_ACTION:
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/STGCN.zip
  batch_size: 1
  max_frames: 50
  display_frames: 80
  coord_size: [384, 512]
  basemode: "skeletonbased"
  enable: False

ID_BASED_DETACTION:
  model_dir:  C:\Users\boyif\Desktop\paddle\football\model\7-16
  batch_size: 1
  basemode: "idbased"
  threshold: 0.5
  display_frames: 80
  skip_frame_num: 2
  enable: True

ID_BASED_CLSACTION:
  model_dir: https://bj.bcebos.com/v1/paddledet/models/pipeline/PPHGNet_tiny_calling_halfbody.zip
  batch_size: 8
  basemode: "idbased"
  threshold: 0.8
  display_frames: 80
  skip_frame_num: 2
  enable: False

REID:
  model_dir:  https://bj.bcebos.com/v1/paddledet/models/pipeline/reid_model.zip
  batch_size: 16
  basemode: "idbased"
  enable: False

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

• [x] Yes I'd like to help by submitting a PR!

[nemonameless]

https://github.com/PaddlePaddle/PaddleDetection/pull/6466

[KirigiriSuzumiya]

6466

mot追踪的问题修复了

• 但是ID_BASED_DETACTION也就是原香烟识别的问题没有解决（检出行为仍显示smoking）

• 且替换模型之后所有人的标注框全部都显示smoking（这个也有可能是我模型的问题，待确认）

• 还有就是能请问一下替换这个香烟识别的模型是一定要有两类的检测模型吗，使用只有一个类的检测模型会报错。

• 如果是一定要两类的话，空类的标注要框什么内容呢

[zoooo0820]

将检测模型的结果转化为行为识别结果的后处理过程请参考这里，检测模型不限类别，需要根据模型情况，将检测结果对应转为行为识别结果； 此外关于对应行为类型的展示，请对应修改此处

[KirigiriSuzumiya]

谢谢，已解决