search

abhyantrika / nanonets_object_tracking

https://nanonets.com/blog/object-tracking-deepsort/
211 stars 118 forks source link

detection file generation and what does each column signify? #9

Open sbanerj2 opened 4 years ago

sbanerj2 commented 4 years ago

Hi, I have q question regarding the detection files from YOLO, Fast-RCNN. How was it generated? As for example in det_yolo3.txt, 1,-1,1094.277,257.304,63.898,62.347,0.400,-1,-1,-1, what does -1 signify? I am guessing that "1" signifies the trackID and the floating numbers following "1" signify the positions of the bounding box and the confidence score. Where is the frame number, class? I have trained my dataset on yolo and was hoping to use the code for tracking. Your help in this regard is really appreciated

Chiayen0503 commented 4 years ago

HI, I have the same issue.

So basically, you train a siamese NN and get a model.pt file. Then you generated object bbox information (txt) produced from object detection model, replace the output with det/*txt. Finally modify ckpt path in test_on_video.py, am I right?

I'm not quite sure how you train and make prediction with this model on a custom data.

Ujang24 commented 4 years ago

Hi, I have q question regarding the detection files from YOLO, Fast-RCNN. How was it generated? As for example in det_yolo3.txt, 1,-1,1094.277,257.304,63.898,62.347,0.400,-1,-1,-1, what does -1 signify? I am guessing that "1" signifies the trackID and the floating numbers following "1" signify the positions of the bounding box and the confidence score. Where is the frame number, class? I have trained my dataset on yolo and was hoping to use the code for tracking. Your help in this regard is really appreciated

The first digit is frame number The second digit is the object ID. -1 means it is not assigned yet. It will be assigned after executing deep_sort The four floating numbers are the box coordinate and size The 0.400 is the confidence score. Don't bother about the last triple -1.

This is the more complete answer: https://github.com/abhyantrika/nanonets_object_tracking/issues/4#issuecomment-586851277