lijun-yu / zero_virus Goto Github PK

Hi Yu,

Thanks for your great work,
I wonder if you use Detectron2 as a detector, it maybe count other classes (instead of car and truck),
In detail, i want to count class Motorbike, bicycle and Bus.
Can i do that?

Hi Yu,

Thanks for your great work,
Can you tell me more detail about the flow of how to set up for a new dataset to run your repo to get the submission?
Thank you!

Hi Lijun Yu, thanks for your great work. Your methods of tracking and routes modeling interested me a lot.

May you tell me, why using features from mask_rcnn_X_101_32x8d for object features extraction task worked really well in case of tracking vehicles? I was really surprised by the tracking quality, because you didn't train and just pick up COCO pre-trained detection model. Is this such as specific of the model?
As i understood, mask_rcnn_X_101_32x8d is mainly COCO detector while solving tracking task is not expected, even that fact the detector was not pretrained for ReID task.

The question could be useful to understand:

1. How to train mask_rcnn_X_101_32x8d to improve tracking (to improve features extraction for ReID)?
2. Could i use another COCO pre-trained detection model (CenterNet, YOLO) and take their features?
3. Or would be better to turn to specific solutions like Joint Detection and Embedding (JDE) models? (In this case JDE-model is usually pre-trained of MOT-datasets with people only, so training is usually needed for vehicle tracking)

Thank you!

Hi team,

Can i run your code to test on my own video?

Thanks!

Hi Yu

as you mentioned in Evaluate, each video has a screenshot of movement and you annotate movements as json file from this.

I wonder how your code can run on private Test set B of AI City challenge, can you are allowed to annotate movement on Test set B?

Is it true that only the online evaluation system will give S1_Effectiveness scores？

*Written in the era of Coronavirus Disease 2019 (COVID-19), with a sincere hope for a better world.

Just a question. Are you alright?
Thanks for your repo.