captaineven / fairmotvehicle

A fork of FairMOT used to do vehicle MOT.用于跟踪车辆的多目标跟踪, 自定义数据进行单类别多目标实时跟踪

Hi there 👋

fairmotvehicle's Issues

what are these models?

from the model_down_script:

ctdet_coco_dla_2x.pth is your baseline fully trained on UA-DETRAC?

What is all_dla34.pth?

Thanks!

problem of 'gen_labels_detrac.py'

gen_dot_train_file(data_root='/home/ubuntu/data/dataset',
rel_path='/DETRAC/images/train',
out_root='/home/ubuntu/FairMOTVehicle/src/data')

clean_train_set(img_root='/home/ubuntu/data/dataset/DETRAC/images/train',
label_root='/home/ubuntu/data/dataset/DETRAC/labels_with_ids/train')

Erase file first and write cfg to get error

Therefore, need to change the position of both functions.

请问将代码修改为多类别多目标跟踪需要修改那一个部分？

您好，
感谢您对MOT任务做出的贡献。我是一个MOT新手，阅读了FairMot源码的issues发现您在下方进行了回复，将源码改成了多类别跟踪。
我想请教下，除了更改update函数以外，还需要更改哪些内容呢？在FairMot源码的基础上修改可以吗？
谢谢，祝好

If I want to run track.py on detrac dataset,how to change the code?

压缩包损坏

代码压缩包下载了好几次显示已损坏

how do i replace tracking id with depth or distance?

how to replace tracking ids with depth information

总loss是负数

Can FairMOT directly detect vehicles without training

Hi !
I am a novice in MOT，I have a question?
Can I directly use the original model(FairMOT) to detect the vehicle?

A problem on training Re-Id heatmap with a vehicle dataset

Hello
How are you?
Thanks for contributing this project.
I am going to train a model for tracking objects like cars.
I have a question about Re-Id.
In general, I think that there may be many "same cars" in a same position or different positions.
Here "same cars" means that they have same shape, color and brand although they are different objects.
In fact, I looked such cases in several vehicle datasets such as KITTI or DETRAC.
We assign different ids to these "same cars" to train Re-Id heatmap.
In this case, can the model for Re-Id converge?
I think that the model may converge in case to train persons rather than cars.
But what about training cars?
I doubt this point.
Thanks

训练detrac数据集的时候验证集曲线出了一些问题

您好，不好意思打扰了。我修改了一下代码，使代码能在每跑完一个epoch后画出验证集的曲线。但是我发现验证集的曲线有一些问题。heatmap的验证集曲线没有下降反而直接增加了。我训练集验证集的比例为8:1。是因为验证集的数据分布和训练集太像了，所以还没有训练完一个epoch就发生了过拟合？我不确定到底是什么问题，您能帮忙解答一下吗？万分感谢。

How did you train the network?

Hi, I'm really interested in fairmot work which is absolutely done perfectly. Could you please share a bit of the detail about how you trained the model for cars? Like did you freeze the body of the network and train only the heads? Did you train id head? How many images did you have for the training data? Did you encounter any generalization errors? I'll be so glad if you can answer my questions.

请问有windows版的吗

terminate called after throwing an instance of 'c10::Error' what(): No CUDA GPUs are available Exception raised from device_count_ensure_non_zero at /pytorch/c10/cuda/CUDAFunctions.cpp:120 (most recent call first):

terminate called after throwing an instance of 'c10::Error'
what(): No CUDA GPUs are available
Exception raised from device_count_ensure_non_zero at /pytorch/c10/cuda/CUDAFunctions.cpp:120 (most recent call first):

这是CUDA的版本问题吗

ua-detrac

hi,
could you show me the file director　of the ua-detrac dataset. I don't know how to arrange the directory of the dataset

No param id.0.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.

你好！
我按照你readme中步骤去训练，报了如下错误：
`/home/ckq/anaconda3/envs/FairMOT/bin/python /home/ckq/git/FairMOTVehicle/src/train.py
Using tensorboardX
Fix size testing.
training chunk_sizes: [1, 2, 1]
The output will be saved to /home/ckq/git/FairMOTVehicle/src/lib/../../exp/mot/default
Setting up data...
Dataset root: /home/ckq/data/even/dataset

dataset summary
OrderedDict([('detrac', 5952.0)])
total # identities: 5952
start index
OrderedDict([('detrac', 0)])

heads: {'hm': 1, 'wh': 2, 'id': 128, 'reg': 2}
opt:
Namespace(K=128, arch='dla_34', batch_size=4, cat_spec_wh=False, chunk_sizes=[1, 2, 1], conf_thres=0.4, data_cfg='../src/lib/cfg/detrac.json', data_dir='/home/ckq/data/even/dataset', dataset='jde', debug_dir='/home/ckq/git/FairMOTVehicle/src/lib/../../exp/mot/default/debug', dense_wh=False, det_thres=0.3, down_ratio=4, exp_dir='/home/ckq/git/FairMOTVehicle/src/lib/../../exp/mot', exp_id='default', fix_res=True, gpus=[0, 5, 6], gpus_str='0, 5, 6', head_conv=256, heads={'hm': 1, 'wh': 2, 'id': 128, 'reg': 2}, hide_data_time=False, hm_weight=1, id_loss='ce', id_weight=1, img_size=(1088, 608), input_h=1088, input_res=1088, input_video='../videos/test3.mp4', input_w=608, is_debug=True, keep_res=False, load_model='../models/ctdet_coco_dla_2x.pth', lr=0.0001, lr_step=[20, 27], master_batch_size=1, mean=None, metric='loss', min_box_area=200, mse_loss=False, nID=5952, nms_thres=0.4, norm_wh=False, not_cuda_benchmark=False, not_prefetch_test=False, not_reg_offset=False, num_classes=1, num_epochs=30, num_iters=-1, num_stacks=1, num_workers=8, off_weight=1, output_format='video', output_h=272, output_res=272, output_root='../results', output_w=152, pad=31, print_iter=0, reg_loss='l1', reg_offset=True, reid_dim=128, resume=False, root_dir='/home/ckq/git/FairMOTVehicle/src/lib/../..', save_all=False, save_dir='/home/ckq/git/FairMOTVehicle/src/lib/../../exp/mot/default', seed=317, std=None, task='mot', test=False, test_mot15=False, test_mot16=False, test_mot17=False, test_mot20=False, track_buffer=30, trainval=False, val_intervals=10, val_mot15=False, val_mot16=False, val_mot17=False, val_mot20=False, vis_thresh=0.5, wh_weight=0.1)
Creating model...
loaded ../models/ctdet_coco_dla_2x.pth, epoch 230
Skip loading parameter hm.2.weight, required shapetorch.Size([1, 256, 1, 1]), loaded shapetorch.Size([80, 256, 1, 1]). If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Skip loading parameter hm.2.bias, required shapetorch.Size([1]), loaded shapetorch.Size([80]). If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
No param id.0.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
No param id.0.bias.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
No param id.2.weight.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
No param id.2.bias.If you see this, your model does not fully load the pre-trained weight. Please make sure you have correctly specified --arch xxx or set the correct --num_classes for your own dataset.
Starting training...`

请问我是哪里错了？

num_class

你好，请问下如果我直接用ＤＥＴＲＡＣ数据集来训练模型的话，ｎｕｍ_class是设置为1还是4呢？我看数据集官网是细分了4中车辆类别，不过我最后只想把训练好的模型用来做车辆跟踪，不需要区分具体的车型，这种情况的话是不是num_class=1就可以了呢？
另一方面我是考虑如果直接设置为1是不是会影响模型训练的效果呢？我看label文件中类别id是从0到3的，所以能否给个建议？
感谢！

sh: 0: Can't open model_down_script.sh

请问我这个训练的结果正常吗？

我看loss都变成负数了

CityPerson.zip is corrupted

Dear author,

The file downloaded from google drive is corrupted.

> unzip Citypersons.zip
error: invalid zip file with overlapped components (possible zip bomb)

Is this the converted dataset from DETRAC? If so, I can download it from there. If not, could you please share the description of the dataset? Thanks!

The docs.google.com/us?xxx in model_down_script.sh cannot be accessed

(pytorch) root@d04c53c86778:~/workspace/ws_wcc/my_docker/fairMotVehicle/FairMOTVehicle# ./model_down_script.sh
mkdir: cannot create directory 'models': File exists
--2023-10-17 14:28:24-- https://docs.google.com/uc?export=download&id=1pl_-ael8wERdUREEnaIfqOV_VF2bEVRT
Resolving docs.google.com (docs.google.com)... 185.60.216.169, 2a03:2880:f136:83:face:b00c:0:25de
Connecting to docs.google.com (docs.google.com)|185.60.216.169|:443...

I tried accessing the connection directly in the browser and it couldn't be opened either.

UA-DETARC测试集上面的指标

您好，请问一下这个模型训练完后您是否在测试集上测了MOTA等指标
结果如何，能否参考一下多谢！

在这个数据集上训练这个模型大概需要多长时间？

请问一下，你们训练这个模型（车辆）花费了多长时间？我目前在Tesla P100 16G上训练该模型，batch size设置成8，其他默认参数。

Missing label.txt in DETRAC

I am facing this error while attempting to train on DETRAC dataset:

(FairMOT)**********:~/FairMOT_ws/FairMOTVehicle/src$ python3 train.py --task mot --exp_id all_dla34 --gpus 0,1 --batch_size 8 --load_model '../models/all_dla34.pth' --data_cfg '../src/lib/cfg/detrac.json' --num_epochs 10 --lr 1e-5
Using tensorboardX
Fix size testing.
training chunk_sizes: [4, 4]
The output will be saved to  /home/*****/FairMOT_ws/FairMOTVehicle/src/lib/../../exp/mot/all_dla34
Setting up data...
Dataset root: /home/*****/FairMOT_ws/Datasets/
Traceback (most recent call last):
  File "train.py", line 135, in <module>
    run(opt)
  File "train.py", line 48, in run
    transforms=transforms)
  File "/home/*****/FairMOT_ws/FairMOTVehicle/src/lib/datasets/dataset/jde.py", line 460, in __init__
    lb = np.loadtxt(lp)
  File "/home/*****/.conda/envs/FairMOT/lib/python3.7/site-packages/numpy/lib/npyio.py", line 981, in loadtxt
    fh = np.lib._datasource.open(fname, 'rt', encoding=encoding)
  File "/home/*****/.conda/envs/FairMOT/lib/python3.7/site-packages/numpy/lib/_datasource.py", line 269, in open
    return ds.open(path, mode, encoding=encoding, newline=newline)
  File "/home/*****/.conda/envs/FairMOT/lib/python3.7/site-packages/numpy/lib/_datasource.py", line 623, in open
    raise IOError("%s not found." % path)
OSError: /home/*****/FairMOT_ws/Datasets/DETRAC/labels_with_ids/train/MVI_39761/img1/img00542.txt not found.
 

the gen_label_detrac.py maybe has something wrong code!

On line 48 to line 54

The following is my modified code：

                        dst_f_path = os.path.join(dst_img1_dir + y)

                        if not os.path.isfile(dst_f_path):

                            shutil.copy(y_path, dst_img1_dir)

                            print('{} cp to {}'.format(y, dst_img1_dir))

                        else:

                            print('{} already exists.'.format(dst_f_path))

No module named 'dcn_v2'