search

MasterBin-IIAU / UNINEXT

[CVPR'23] Universal Instance Perception as Object Discovery and Retrieval
MIT License
1.5k stars 158 forks source link

MOT on video #8

Closed chema-344 closed 1 year ago

chema-344 commented 1 year ago

Hello,

Great work!!. Could you provide an example of MOT using a video as input?

I'm having trouble trying.

Many thanks

MasterBin-IIAU commented 1 year ago

Hi, we have upload an demo video from BDD100K MOT val to our onedrive. Please download and unzip demo_bdd.zip then you can get a folder with structure datasets/bdd. This folder should contain video frames and the corresponding json file.

Then open your local projects/UNINEXT/uninext/uninext_vid.py then change line 53 to self.debug_only=True and comment out line 396 to line 399.

Finally, run the following command python3 launch.py --nn 1 --np 1 --eval-only --uni 1 --config-file projects/UNINEXT/configs/eval-vid/video_joint_r50_eval_mot.yaml --resume OUTPUT_DIR outputs/video_joint_r50 MODEL.USE_IOU_BRANCH False. Then you should see the detection and tracking results drawn on the original frames. Specifically, detection results starts with frame_det and different colors represent different object classes. Besides, tracking results starts with frame_track and different colors represent different object ID. Enjoy :)

MasterBin-IIAU commented 1 year ago

@chema-344 Hi, have the answer solve your issue? If the issue has been solved, please consider closing it. Thanks :)

chema-344 commented 1 year ago

Thanks!!!

ajdroid commented 1 year ago

Hi @MasterBin-IIAU, in the results I get from following your instructions above, the track frames have different colors between subsequent frames -- why does this happen? Is an image model being run and each frame is initialised separately?

beratersari commented 1 year ago

Hi, we have upload an demo video from BDD100K MOT val to our onedrive. Please download and unzip demo_bdd.zip then you can get a folder with structure datasets/bdd. This folder should contain video frames and the corresponding json file.

Then open your local projects/UNINEXT/uninext/uninext_vid.py then change line 53 to self.debug_only=True and comment out line 396 to line 399.

Finally, run the following command python3 launch.py --nn 1 --np 1 --eval-only --uni 1 --config-file projects/UNINEXT/configs/eval-vid/video_joint_r50_eval_mot.yaml --resume OUTPUT_DIR outputs/video_joint_r50 MODEL.USE_IOU_BRANCH False. Then you should see the detection and tracking results drawn on the original frames. Specifically, detection results starts with frame_det and different colors represent different object classes. Besides, tracking results starts with frame_track and different colors represent different object ID. Enjoy :)

that worked for me but just write the image files with bboxes. How can show it with segmented pixels? Could you help me?