CaptainEven / MCMOT

Real time one-stage multi-class & multi-object tracking based on anchor-free detection and ReID
MIT License
383 stars 82 forks source link

wrong detection for visdron validation dataset #51

Closed tiru1930 closed 3 years ago

tiru1930 commented 3 years ago

Hi I am trying to validate visdron validation dataset but the results, are not yet all good.

python demo.py

test_single(img_path='/home/tiru/Desktop/mcmot/MCMOT/data/VisDrone2019-MOT-val/sequences/uav0000339_00001_v/0000003.jpg', dev=torch.device('cpu'))

pre-trained model - model_path = './models/mcmot_last_track_resdcn_18_visdrone.pth'

0000003

CaptainEven commented 3 years ago

@You need to modify object classes first: image

CaptainEven commented 3 years ago

@tiru1930 Modify object class ids too. image

CaptainEven commented 3 years ago

uav_339_track

tiru1930 commented 3 years ago

@CaptainEven I have done those changes but results are same

        self.parser.add_argument('--reid_cls_ids',
                                 default='0,1,2,3,4,5,6,7,8,9',  # '0,1,2,3,4' or '0,1,2,3,4,5,6,7,8,9'
                                 help='')  # the object classes need to do reid

from lib.utils.post_process import ctdet_post_process
from .basetrack import BaseTrack, MCBaseTrack, TrackState
from gen_dataset_visdrone import cls2id, id2cls  # visdrone
# from gen_labels_detrac_mcmot import cls2id, id2cls  # mcmot_c5

0000003

tiru1930 commented 3 years ago

@CaptainEven can i get the sample video that u are using for testing ?

CaptainEven commented 3 years ago

@tiru1930 Using ffmpeg, you will get the video from images.

tiru1930 commented 3 years ago

@CaptainEven over the frames , track id are changing, is this expected from these models ?

csdingithub commented 2 years ago

@CaptainEven I have done those changes but results are same

        self.parser.add_argument('--reid_cls_ids',
                                 default='0,1,2,3,4,5,6,7,8,9',  # '0,1,2,3,4' or '0,1,2,3,4,5,6,7,8,9'
                                 help='')  # the object classes need to do reid

from lib.utils.post_process import ctdet_post_process
from .basetrack import BaseTrack, MCBaseTrack, TrackState
from gen_dataset_visdrone import cls2id, id2cls  # visdrone
# from gen_labels_detrac_mcmot import cls2id, id2cls  # mcmot_c5

0000003

the same result

csdingithub commented 2 years ago

I solve the problem by editing the code in demo.py!

# line 143/144 # Head dimensions of the net heads = {'hm': 10, 'reg': 2, 'wh': 2, 'id': 128} when I use the visdrone,I also need to change the heads structure!

dragen1860 commented 2 years ago

@csdingithub Hi, I do not understand what you mean: I also need to change the heads structure! please give more details. thank you .