vehicle tracking video not worked

[MyraBaba]

问题确认 Search before asking

• [X] 我已经搜索过问题，但是没有找到解答。I have searched the question and found no related answer.

请提出你的问题 Please ask your question

Hi,

I ran :

python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_ppvehicle.yml \
                                                   --video_file=myVideo.mp4 \
                                                   --device=gpu

without changing anything in infer_cfg_ppvehicle.yml , the resulting video is the same as input video. No boxes and trajectories ? additionally how I can export Box and related vehichle_id ?

Best

[nemonameless]

See doc, please set enable: True if you run as tracking mode. https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/deploy/pipeline/config/infer_cfg_ppvehicle.yml#L14

[MyraBaba]

Thanks :)

I tested a intersection video , my obsevartion is:

Smal vehicle is not detected İf there is an obstacle even small the tracking loosing the object %80

Wha we can do improve the small vehicle detection and also tracking performance ? Any other model ?

Yolov7-e6.pt we had better detection perf.

Best

On 14 Sep 2022, at 19:28, Feng Ni @.***> wrote:

See doc https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/deploy/pipeline/docs/tutorials/ppvehicle_mot_en.md#how-to-use, please set enable: True if you run as tracking mode. https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/deploy/pipeline/config/infer_cfg_ppvehicle.yml#L14 https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.5/deploy/pipeline/config/infer_cfg_ppvehicle.yml#L14 — Reply to this email directly, view it on GitHub https://github.com/PaddlePaddle/PaddleDetection/issues/6939#issuecomment-1247017585, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEFRZH63Z63ACD7Z7NKUXBDV6H4KBANCNFSM6AAAAAAQMQHQLU. You are receiving this because you authored the thread.

[nemonameless]

https://github.com/PaddlePaddle/PaddleDetection/tree/develop/configs/ppvehicle The models are trained on BDD100K and UA-DETRAC, the vehicle scene may be very different from your scene, you need training or finetuning for your own dataset. Have you ever trained your dataset on Yolov7-e6? Maybe you can directly try COCO weights, see ppyoloe plus

[MyraBaba]

Not specially trained with yolov7-e6 , I tested as its vanilla model.

[MyraBaba]

How we push the gpu to %100 ? infer.py and mot_tracking only using %33 gpu