Detecting Traffic Lights by Single Shot Detection

This video shows traffic light detection results using the Single Shot Detector (SSD) method and an Inception-v3 base network. Several adaptions were done to detect small objects. Details see in the paper. The DriveU Traffic Light Dataset (DTLD) was used for training and testing.
Paper: arxiv.org/abs/...
Github soon: github.com/jul...
Dataset: traffic-light-d...

[@mohammadraselamin1473]

good work, i request for dataset download, please send the link

[@Shuji_Ono_Xin1]

This is a great video. Does this process work in real time?

[@JM-jk6jt]

Thanks Shuji. Yes, Processing time is 60 miliseconds on a Nvidia Titan X

[@srinidhid4127]

Hi, great work! I have registered and downloaded the dataset. But after downloading and extracting with 7-zip file still I cannot see the images it is only black! but the disparity images were black and white. How can this be resolved? Thanks

[@JM-jk6jt]

github.com/julimueller/dtld_parsing/issues/2

[@Engro39]

thank you for showing the result video. is any tracker applied after the detection?

[@JM-jk6jt]

Nope, just single frame detection.

[@sriaravind9468]

Really great work! But the dataset link is not working, could you provide an alternate one?

[@JM-jk6jt]

Hi Sri, thanks and sorry for the issues you are facing. I will check this. In the meantime, I suggest to use the direct link: www.uni-ulm.de/en/in/driveu/projects/driveu-traffic-light-dataset/

[@nishi7274]

hi ,I can't access the dataset , how many days it take to approve the dataset access ?

[@JM-jk6jt]

I can see your registration. Confirmation and Download URL is immediately sent out, so you should have received it already. Can you check your spam folder?

[@nishi7274]

@@JM-jk6jt Thank you 😊