FairMOT
FairMOT copied to clipboard
ifzhang
[IJCV-2021] FairMOT: On the Fairness of Detection and Re-Identification in Multi-Object Tracking
First of all, FairMOT is a very good job, very good at tracking effectiveness and efficiency. I want to apply this work with cross-camera tracking, but need to verify the...
I used the model mot20_fairmot.pth given by the author to upload to the MOT20 test set to see the results. The results in the test set cannot reach the accuracy...
How can I solve the below problem? ########## Train Log ############### mot/crowdhuman_dla34| train: [1][4240/19370]|Tot: 0:28:23 |ETA: 0:00:00 |loss 20.8643 |hm_loss 1.6064 |wh_loss 4.8034 |off_loss 0.2379 |id_loss 13.2775 |Data 0.020s(0.020s) |Net...
我执行了指令： python track.py mot --data_cfg ../src/lib/cfg/mot17.json --data_dir /my_tmp/zwh/Dataset_src/MOT17/ --test_mot17 True --load_model ../models/fairmot_dla34.pth --conf_thres 0.4 主要修改了数据集的路径，但是得到结果如下： IDF1 IDP IDR Rcll Prcn GT MT PT ML FP FN IDs FM MOTA MOTP...
When I try using demo.py for my own video to test tracking, I notice that there is: Lenth of the video: 0 frames but the processing does not stop yet,...
id_head = self.emb_scale * F.normalize(id_head) So why do you use emb_scale?
 last_index已经是id总数了，为什么还要加一呢，有什么含义呢
C:\ProgramData\Anaconda3\envs\pytorch_fair\python.exe F:/code/py/FairMOT-master/src/track.py mot --load_model ../models/fairmot_dla34.pth --val_mot16 Ture --conf_thres 0.4 2022-06-23 10:04:53 [INFO]: start seq: MOT17-02-SDP Fix size testing. training chunk_sizes: [6, 6] The output will be saved to F:\code\py\FairMOT-master\src\lib\..\..\exp\mot\default heads...
Dear all: I noticed that the cross entropy ID loss for small ID number dataset works pretty well. But when scaled to large-scale ID numbers, let's say up to 0.1M...
