search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.58k stars 7.95k forks source link

question about object re identification from two cameras #4572

Open PROGRAMMINGENGINEER-NIKI opened 4 years ago

PROGRAMMINGENGINEER-NIKI commented 4 years ago

Hello @AlexeyAB

Thank you for your effort in making this repository.

I've been recently using the YOLO to detect trucks in images, which turned out really well. My next step is to try to find images of the same truck across the whole set of images I've retrieved earlier. To me, the input of the system is an image and its outputs are k candidate images, ranked accordingly to the system's confidences. In other words, the system will try to match an image with the other images he works with.

My question is how can one reidentify trucks after detecting them through the YOLO network? Please advise! Thanks

xrv0 commented 4 years ago

I dont think this is possible without adding some functionality yourself. You cant compare two images by comparing the confidence factor because this will vary by angel, light etc. You could try to extract the bbox and compare them to others which you previous saved.

AlexeyAB commented 4 years ago

@PROGRAMMINGENGINEER-NIKI If you want to compare images which are not the sequence frames from 1 video, then you should use Siamese networks to compare trucks, which are not implemented in the Darknet.

Otherwise, if you use 1 video (file/camera) then you can train https://github.com/AlexeyAB/darknet/files/3199770/yolo_v3_tiny_lstm.cfg.txt model as described there https://github.com/AlexeyAB/darknet/issues/3114#issuecomment-494154586

Un-comment this line: https://github.com/AlexeyAB/darknet/blob/master/src/yolo_console_dll.cpp#L16

Compile Darknet with LIBSO=1 in the Makefile

And run detection: LD_LIBRARY_PATH=./:$LD_LIBRARY_PATH ./uselib data/coco.names cfg/yolov3.cfg yolov3.weights test.mp4

So there will be Detection (Yolo) and Tracking (Optical flow), so each object will have unique trakc_id.

RollingIsland commented 4 years ago

Hi, @AlexeyAB . I have downloaded yolo_v3_tiny_pan_lstm.cfg.txt and weights-file. I wonder that whether the .name file should be changed or not, and can it work with yolo_cpp_dll directly?

AlexeyAB commented 4 years ago

can it work with yolo_cpp_dll directly?

Yes.

.name file should be changed

It depends on names of your objects

RollingIsland commented 4 years ago

@AlexeyAB , Thanks for your quick reply! Well, the weights file is downloaded from the link you offered and I have no idea what objects it was trained with... I have tried to use coco.names directly but it detect cars as "person".