使用DeepSORT模型评估，采用经过匹配之后的检测框结果det_results_dir报错

[ilovefanjian]

问题确认 Search before asking

• 我已经查询历史issue，没有发现相似的bug。I have searched the issues and found no similar bug report.

Bug组件 Bug Component

No response

Bug描述 Describe the Bug

使用python tools/eval_mot.py -c configs/mot/deepsort/reid/deepsort_pcb_pyramid_r101.yml --det_results_dir det_results_dir加载det_results评估报错：
W1211 16:45:02.119659 12084 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 8.6, Driver API Version: 12.5, Runtime API Version: 11.7
W1211 16:45:02.123663 12084 gpu_resources.cc:91] device: 0, cuDNN Version: 8.4.
[12/11 16:45:02] ppdet.data.source.category WARNING: anno_file 'None' is None or not set or not exist, please recheck TrainDataset/EvalDataset/TestDataset.anno_path, otherwise the default categories will be used by metric_type.
[12/11 16:45:02] ppdet.data.source.category WARNING: metric_type: MOT, load default categories of pedestrian MOT.
[12/11 16:45:02] ppdet.utils.checkpoint INFO: Finish resuming model weights: C:\Users\whkedx2023/.cache/paddle/weights\deepsort_pplcnet.pdparams
[12/11 16:45:02] ppdet.engine.tracker INFO: Evaluate seq: MOT16-02
[12/11 16:45:02] ppdet.engine.tracker INFO: Found 600 inference images in total.
[12/11 16:45:03] ppdet.engine.tracker INFO: Finish loading detection results file dataset/mot/det_results_dir\MOT16-02.txt.
0%|▎ | 1/600 [00:02<23:52, 2.39s/it]
Traceback (most recent call last):
File "tools/eval_mot.py", line 144, in
main()
File "tools/eval_mot.py", line 140, in main
run(FLAGS, cfg)
File "tools/eval_mot.py", line 99, in run
det_results_dir=FLAGS.det_results_dir)
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\engine\tracker.py", line 548, in mot_evaluate
'{}.txt'.format(seq)))
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\engine\tracker.py", line 379, in _eval_seq_sde
online_targets = tracker.update(pred_dets, pred_embs)
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\modeling\mot\tracker\deepsort_tracker.py", line 121, in update
File "ppdet\modeling\mot\motion\kalman_filter.py", line 29:
def nb_project(mean, covariance, std, _update_mat):

mean = np.dot(_update_mat, mean)
covariance = np.dot(np.dot(_update_mat, covariance), _update_mat.T)
^

复现环境 Environment

Windows，PaddleDetection: release/2.6， python: 3.8

Bug描述确认 Bug description confirmation

• 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!

[Bobholamovic]

请问方便贴一下完整的信息嘛？看起来似乎缺少了最后一段

[ilovefanjian]

完整信息

[Bobholamovic]

看起来可能是一个数据类型问题，建议尝试修改代码，将报错位置改成：

covariance = np.dot(np.dot(_update_mat.astype(np.float32), covariance.astype(np.float32)), _update_mat.T.astype(np.float32))

[ilovefanjian]

又出现了新的错误
(PaddleDetection) D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6>python tools/eval_mot.py -c configs/mot/deepsort/reid/deepsort_pcb_pyramid_r101.yml --det_results_dir dataset/mot/det_results_dir
W1212 16:36:55.255107 728 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 8.6, Driver API Version: 12.5, Runtime API Version: 11.7
W1212 16:36:55.258111 728 gpu_resources.cc:91] device: 0, cuDNN Version: 8.4.
[12/12 16:36:55] ppdet.data.source.category WARNING: anno_file 'None' is None or not set or not exist, please recheck TrainDataset/EvalDataset/TestDataset.anno_path, otherwise the default categories will be used by metric_type.
[12/12 16:36:55] ppdet.data.source.category WARNING: metric_type: MOT, load default categories of pedestrian MOT.
[12/12 16:36:56] ppdet.utils.checkpoint INFO: Finish resuming model weights: mot_det_weight/deepsort_pcb_pyramid_r101.pdparams
[12/12 16:36:56] ppdet.engine.tracker INFO: Evaluate seq: MOT16-02
[12/12 16:36:56] ppdet.engine.tracker INFO: Found 600 inference images in total.
[12/12 16:36:57] ppdet.engine.tracker INFO: Finish loading detection results file dataset/mot/det_results_dir\MOT16-02.txt.
0%|▎ | 1/600 [00:03<33:53, 3.40s/it]
Traceback (most recent call last):
File "tools/eval_mot.py", line 144, in
main()
File "tools/eval_mot.py", line 140, in main
run(FLAGS, cfg)
File "tools/eval_mot.py", line 99, in run
det_results_dir=FLAGS.det_results_dir)
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\engine\tracker.py", line 548, in mot_evaluate
'{}.txt'.format(seq)))
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\engine\tracker.py", line 379, in _eval_seq_sde
online_targets = tracker.update(pred_dets, pred_embs)
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\modeling\mot\tracker\deepsort_tracker.py", line 121, in update
detections[detection_idx])
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\modeling\mot\tracker\base_sde_tracker.py", line 127, in update
detection.to_xyah())
File "D:\DeepLearning\two\hzhlearning\PaddleDetection-release-2.6\ppdet\modeling\mot\motion\kalman_filter.py", line 262, in update
measurement, self._update_mat)
File "D:\DeepLearning\Anaconda3\envs\PaddleDetection\lib\site-packages\numba\core\dispatcher.py", line 468, in _compile_for_args
error_rewrite(e, 'typing')
File "D:\DeepLearning\Anaconda3\envs\PaddleDetection\lib\site-packages\numba\core\dispatcher.py", line 409, in error_rewrite
raise e.with_traceback(None)

File "ppdet\modeling\mot\motion\kalman_filter.py", line 43:
def nb_update(mean, covariance, proj_mean, proj_cov, measurement, meas_mat):
kalman_gain = np.linalg.solve(proj_cov, (covariance @meas_mat.T).T).T

[Bobholamovic]

错误信息似乎不完整……如果还是类似的报错的话，可以采用类似的改写方法，即添加.astype(np.float32)。

[ilovefanjian]

谢谢，问题已经解决，请问bytetrack和其他算法，如何通过加载检测框结果det_results_dir的方式在自制数据集上进行跟踪评估。

[Sunting78]

您好 use_detector设置为False 并且传入det_file 相关接口：

PaddleDetection/ppdet/engine/tracker.py

Line 244 in b972633

def _eval_seq_sde(self,