search

Robootx / Multi-Camera-Object-Tracking-via-Transferring-Representation-to-Top-View

Multi-Camera Object Tracking via Deep Metric Learning
186 stars 64 forks source link

Real time detection #10

Open Knightfire1998 opened 4 years ago

Knightfire1998 commented 4 years ago

hello i want to use multiple webcams as input how can i do that i tried adding ip adrresses in the cap list but it didnt work please help thank you

abnercloud commented 3 years ago

same question

HoiBunCa commented 3 years ago

I have an idea like this: you can use yolov5 to get the frame and bounding box of the object in that frame. Then you pass boxes and frames to the get_embedding function and run as usual. I tried it and got it running. For some corporate security reasons, I cannot publish the code for you, but if you have any questions, please ask me questions, I will help you as much as I can.

HoiBunCa commented 3 years ago

bonus: you can pass the camera's ip to the Loadstream class of yolov5 as input

sry948749099 commented 3 years ago

bonus: you can pass the camera's ip to the Loadstream class of yolov5 as input

it is taking too much minutes to complete the whole process,How did you solve this problem? If i want to run it with my own dataset, does it just need to be modified the .pickle file corresponding to the video?

Yingjie1999 commented 3 years ago

same question hello Have you solved it yet please

sry948749099 commented 3 years ago

same question hello Have you solved it yet please

I tested on my own dataset, but the results are very terrible, I don't know what the problem is. if you want to communicate more, please contact by email: 1605938502@qq.com

Yingjie1999 commented 3 years ago

I have an idea like this: you can use yolov5 to get the frame and bounding box of the object in that frame. Then you pass boxes and frames to the get_embedding function and run as usual. I tried it and got it running. For some corporate security reasons, I cannot publish the code for you, but if you have any questions, please ask me questions, I will help you as much as I can.

Hello, after I get the .pickle file, I use my own video, but the recognition effect is not good, it often identifies several people in the video as one person. In addition, I used its Campus sequences from https://www.epfl.ch/labs/cvlab/data/data-pom-index-php/ (official）, but the effect was not good. Do you know why？please help, thank you