Demo

[iazdan]

Hi,
Thanks for your wonderful work. Would that be possible to add a demo.py script for testing videos?

Thanks

[SkalskiP]

Hi 👋🏻 Google Colab showing how to set up and run the tracker would be awesome as well.

[SangbumChoi]

@iazdan @SkalskiP #11 (comment)
Enjoy and tell me if there is something wrong

[GuHuangAI]

@SangbumChoi
Excellent work! We would like to check the code and merge soon.

[SkalskiP]

@SangbumChoi The one problem I see is that DiffMOT is currently tightly entangled with YOLOX. In theory, it could work with any detector, but there is no easy way to do it currently.

For example, I'd like to plug DiffMOT into my demo project here: https://x.com/skalskip92/status/1788253029965140396, where I do not use YOLOX and for now, I have no idea how to do it. :) I'd appreciate any help. 🙏🏻

[panwpalo]

Thank you @SangbumChoi for your quick response. I get the error below while executing the demo.py file. Would you mind fixing this issue as well?

File "demo.py", line 10, in
from diffmot import DiffMOT
File "DiffMOT/diffmot.py", line 13, in
from models.autoencoder import D2MP
ModuleNotFoundError: No module named 'models.autoencoder'

[SangbumChoi]

@panwpalo that is the installation error not the code part. please follow the basic installation part.

[SangbumChoi]

@SkalskiP That is true, you do not need to have YOLOX extension. You only need is detection result prepared as follow
<frame>, <x>, <y>, <x2>, <y2>, <confidence>
with the txt file in the dataset_dir.

Do you have the mp4 and the result txt file?

[SangbumChoi]

@GuHuangAI FYI, I only tested in DanceTrack and I think we need some further modification for other weight file too

[SkalskiP]

@SangbumChoi yes, I have both files. What should I do now?

[SangbumChoi]

@SkalskiP Could you share here? I can modify and add parser to the code to make it compatible

[MuhammadRamzan4]

@SangbumChoi Try to use your demo.py file but facing this error

[SangbumChoi]

@MuhammadRamzan4 I think you do not have image files path.

[panwpalo]

After running the demo.py the only output that I see is the video being converted into images and it gets stuck afterward! Any clue ?

2024-05-10 16:09:20 [DEBUG]: matplotlib data path: /opt/anaconda3/envs/diffmot/lib/python3.9/site-packages/matplotlib/mpl-data
2024-05-10 16:09:20 [DEBUG]: CONFIGDIR=/.matplotlib
2024-05-10 16:09:20 [DEBUG]: interactive is False
2024-05-10 16:09:20 [DEBUG]: platform is darwin
2024-05-10 16:09:20 [DEBUG]: CACHEDIR=.matplotlib
2024-05-10 16:09:20 [DEBUG]: Using fontManager instance from.matplotlib/fontlist-v330.json

Train Dataset built!
File 'external/weights/dance_sbs_S50.pth' already exists. Skipping download.
File 'experiments/mot_ddm_1000_deeper/DanceTrack_epoch800.pt' already exists. Skipping download.
File 'weights/DanceTrack_yolox_s.tar' already exists. Skipping download.

[SangbumChoi]

@panwpalo You should wait until the YOLO process, how long is your video size? usually 300 frames gets done in a 10 seonds in good GPU

[cedricnam]

@SangbumChoi i got error

[panwpalo]

I used the default video and yolox and I am running it on cpu. Is the mmtracking implementation faster than this Repo?

[SangbumChoi]

@panwpalo I think it is due to using cpu use light yolo model

[Coronal-Halo]

In the "compute_embedding(self, img, bbox, tag)" function, when you clip the resulting bounding boxes, the 'results' may end up being empty, as shown in the above image (both p[0] and p[2] are 0 due to clipping -- they were originally negative). In this case, the empty 'results' variable will cause errors in the following code (for example, in the next line 135: crop = cv2.cvtColor(crop, cv2.COLOR_BGR2RGB), since cv2.cvtColor does not accept empty array). How do you resolve this bug ?

[Abhipray04]

@SangbumChoi Thanks for doing this! It helped a lot.

A quick question - how can I make it work for SportsMOT instead of DanceTrack? What modifications do I need to make to use SportsMOT weights?

Thanks again.

[Rainbow5495]

@SangbumChoi Try to use your demo.py file but facing this error

I have the same problem. Have you solved it?

[Rainbow5495]

@panwpalo I think it is due to using cpu use light yolo model

Could you please offer me some help? Thanks a lot!