How would you change tracking from people to cars? about deep_sort_yolov3 HOT 14 CLOSED

yolo is used for detection, which can detect 80 classes with coco dataset.
deep_sort can only calculate person appearance features with default model. even though you detect multi targets such as cars, truck, bird etc with yolo, the deep_sort tracking performace would not be well.

you need train your own model for tracking.

from deep_sort_yolov3.

Change the part in the yolo detector code where we filter 'person' from detected boxes.
In yolo.py

for i, c in reversed(list(enumerate(out_classes))):
predicted_class = self.class_names[c]
if predicted_class != 'person' :
continue
box = out_boxes[i]
Change 'person' to 'car'.

from deep_sort_yolov3.

Excellent! Missed that, thank you for the pointer! @anshu1106

from deep_sort_yolov3.

Maybe not , the deep_sort model was just trained for person

from deep_sort_yolov3.

I can confirm that it tracks cars, birds and trucks too and performs well.

from deep_sort_yolov3.

Maybe deep_sort more like a feature coding and similarity comparator , no matter what object it is .

from deep_sort_yolov3.

hi!, there's a way to track multiple objects?, like cars and trucks at the same time?

from deep_sort_yolov3.

you can have a try . they all belong to "car" , just like @anshu1106 say to have a try .

from deep_sort_yolov3.

Thanks a lot ,btw it can be easily used to track cars. It's favoured if the relevant paper can be provide : )

from deep_sort_yolov3.

I want to track other things, not people, cars, birds, etc. However, the tracking effect is not good. How should I train again?

from deep_sort_yolov3.

@Goldwh You can reference to https://github.com/nwojke/deep_sort

from deep_sort_yolov3.

hi!, there's a way to track multiple objects?, like cars and trucks at the same time?

yes.

Change the part in the yolo detector code where we filter 'person' from detected boxes.
In yolo.py

for i, c in reversed(list(enumerate(out_classes))):
predicted_class = self.class_names[c]
if predicted_class != 'car' and predicted_class!='truck' :
continue
box = out_boxes[i]
Change 'person' to 'car'.

from deep_sort_yolov3.

hi! I ran demp.py. than this error occur
VIDEOIO ERROR: V4L2: Pixel format of incoming image is unsupported by OpenCV
can anyone help me figure out some way to overcome this error?

from deep_sort_yolov3.

yolo is used for detection, which can detect 80 classes with coco dataset.
deep_sort can only calculate person appearance features with default model. even though you detect multi targets such as cars, truck, bird etc with yolo, the deep_sort tracking performace would not be well.

you need train your own model for tracking.

I agree. DeepSort is trained for Person Re-Identification. I wanted to track people and cars together, I observed a lot of ID switches for vehicles. so, If you want to use the same model to track multiple objects then it would need to be trained again.

from deep_sort_yolov3.