search

ifzhang / FairMOT

[IJCV-2021] FairMOT: On the Fairness of Detection and Re-Identification in Multi-Object Tracking
MIT License
4.01k stars 932 forks source link

Train on custom dataset #254

Open nwoyecid opened 4 years ago

nwoyecid commented 4 years ago

What is your dataset format in case one want to train on a custom dataset. I have hard time understanding the format of the data passed to your data loader. In your data directory, I can only see list of image paths. How (format) and where did you put the labels. Can you examplify this (maybe on a random data)?

ifzhang commented 4 years ago

I have updated some instructions for training on custom dataset. You can find it in ReadMe.

nwoyecid commented 4 years ago

Thanks for your response. Can this work for multi-class multi-object tracking. Is it straightforward in your code to make some changes and be able to track both multiple humans, multiple cars, multiple dogs, etc?

CaptainEven commented 4 years ago

@nwoyecid You can refer to this repo https://github.com/CaptainEven/MCMOT, which is an extention work of FairMOT.

nwoyecid commented 4 years ago

Thanks, I am having a little hard time understanding your track.py file. Would you mind providing some descriptions to the arguments passed to the def main(args) function. I don't understand them enough to translate to my custom dataset.

Also, what is img1 in this line of code? dataloader = datasets.LoadImages(osp.join(data_root, seq, 'img1'), opt.img_size)

kevinwei30 commented 4 years ago

I think the img1 is only for the MOT dataset, because of its default file structure. For me, to use custom dataset, just create my own dataloader and then use eval_seq().

geyu1998 commented 3 years ago

when i train on custom dataset,there is a error. IsADirectoryError: [Errno 21] Is a directory: '/FairMOT-master/datas/' i put my dataset in datas file. i don't know why ,how can i solve this problem