MedChaabane / DEFT

Joint detection and tracking model named DEFT, or ``Detection Embeddings for Tracking." Our approach relies on an appearance-based object matching network jointly-learned with an underlying object detection network. An LSTM is also added to capture motion constraints.
MIT License
269 stars 43 forks source link

Why the test result of KITTI only include car class? #10

Closed zhangchuang-zc closed 3 years ago

zhangchuang-zc commented 3 years ago

HI! Thanks for your excellent work! I run the test.py to get the KITTI result, but it only includes car class,not pedestrians and cyclist. Does your code only focus the car class during training, or does it only detect the car class during testing? I don't find the corresponding code in the project, can you help me? @MedChaabane Thanks!

MedChaabane commented 3 years ago

Hi, Thanks for your interest in our work! In KITTI we just trained and tested our model on car class. in nuScenes dataset, we have tested our model for tracking car, pedestrians, cyclist and also four other classes. Thanks,

zhangchuang-zc commented 3 years ago

Thanks for your reply! How did you achieve only tracking car in the training and testing process?I found your model model_kitti.pth corresponds to three classes, but the testing results only include car class. I did not find the corresponding code in your project, can you help me find the corresponding code? I want to train a model that only tracks pedestrians, can you help me? @MedChaabane Thank you very much!

MedChaabane commented 3 years ago

Hi, in the tracker code (https://github.com/MedChaabane/DEFT/blob/main/src/lib/utils/tracker.py ) , line 793 we filter only car class for tracking, " if det["class"] == 2 " . You can change in that line to the class id of pedestrian instead of class id of car "2" and it should test on pedestrians. However, the kitti train data does not have enough data on pedestrian class so I don't expect that it will works well on pedestrians without pre-training it on other pedestrian datasets first.

zhangchuang-zc commented 3 years ago

Thanks for your help! @MedChaabane