problems in testing

[gitac10]

when i use the model to testing. if i use the python 3.6, it shows

Traceback (most recent call last):
  File "tools/test.py", line 22, in <module>
    import mmdet_custom  # noqa: F401,F403
  File "/data/liuyida/liuyida/mmped/MMPedestron-main/mmdet_custom/__init__.py", line 7, in <module>
    from .models import *  # noqa: F401,F403
  File "/data/liuyida/liuyida/mmped/MMPedestron-main/mmdet_custom/models/__init__.py", line 9, in <module>
    from .utils import *  # noqa: F401,F403
  File "/data/liuyida/liuyida/mmped/MMPedestron-main/mmdet_custom/models/utils/__init__.py", line 5, in <module>
    from .transformer import DetrTransformerEncoder_CP
  File "/data/liuyida/liuyida/mmped/MMPedestron-main/mmdet_custom/models/utils/transformer.py", line 5, in <module>
    from fairscale.nn.checkpoint import checkpoint_wrapper
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/__init__.py", line 12, in <module>
    from . import nn
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/nn/__init__.py", line 9, in <module>
    from .data_parallel import FullyShardedDataParallel, ShardedDataParallel
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/nn/data_parallel/__init__.py", line 8, in <module>
    from .fully_sharded_data_parallel import FullyShardedDataParallel, OffloadConfig, TrainingState, auto_wrap_bn
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/nn/data_parallel/fully_sharded_data_parallel.py", line 43, in <module>
    from fairscale.nn.misc import FlattenParamsWrapper
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/nn/misc/__init__.py", line 12, in <module>
    from .flatten_params_wrapper import FlattenParamsWrapper
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/nn/misc/flatten_params_wrapper.py", line 34, in <module>
    from fairscale.experimental.nn.ssd_offload import SsdFlatParameter
  File "/data/liuyida/liuyida/anaconda3/envs/mmped/lib/python3.6/site-packages/fairscale/experimental/nn/ssd_offload.py", line 6
    from __future__ import annotations
    ^

so i change the python version to python 3.7. but then I encountered such a problem.

  File "/data/liuyida/liuyida/anaconda3/envs/myped37/lib/python3.7/importlib/__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
ImportError: libcudart.so.11.0: cannot open shared object file: No such file or directory

[PredelinaAsya]

@gitac10 Good afternoon! Tell me, please, how did you fix this error? Faced the same problem

[gitac10]

@gitac10 Good afternoon! Tell me, please, how did you fix this error? Faced the same problem

I switch to python 3.7 and then reinstall cuda11.0. but it seems i meet other problems。 you can refer this，It seems that the CUDA 11.0 environment is missing. https://blog.csdn.net/hai4321/article/details/133916420 Is it the former case? At present, I haven't solved it either. If you solve it, can you tell me?

[PredelinaAsya]

It turned out to fix a bug with python=3.6 installing a older version of fairscale==0.4.0. But after that, a new problem arose: ModuleNotFoundError: No module named 'petrel_client'. You wrote about it in the issue

[gitac10]

It turned out to fix a bug with python=3.6 installing a older version of fairscale==0.4.0. But after that, a new problem arose: ModuleNotFoundError: No module named 'petrel_client'. You wrote about it in the issue

it seems i solve the petrel_client error by modifying the config py file_client_args=dict(backend='disk')), try change the backend=petrel_client to backend=disk You can see the explanation in the file_client.py file.

[PredelinaAsya]

Thank you! I also fixed a bug with petrel_client. Now there are problems with uploading the dataset. Did you manage to run the pipeline after correcting this error?

[gitac10]

Thank you! I also fixed a bug with petrel_client. Now there are problems with uploading the dataset. Did you manage to run the pipeline after correcting this error?

seems i successfully run the code. i use the eventped dataset, and modified the mmpedestron_mix5datasets_best.py, get the results corresponding to the paper.i just change the dataset path,and the weight path

[PredelinaAsya]

@gitac10 Good afternoon! Tell me, please, have you run model on any other datasets? I have a problem: when running on Crowdhuman validation, metrics are obtained that are much lower than in the paper. If you have run the model on Crowdhuman, please share your metrics and config

[gitac10]

@gitac10 Good afternoon! Tell me, please, have you run model on any other datasets? I have a problem: when running on Crowdhuman validation, metrics are obtained that are much lower than in the paper. If you have run the model on Crowdhuman, please share your metrics and config

good morning. i did't test the crowdhuman validation. maybe you can change the weight and try? as for me ,i test the eventped dataset , it's correspond with the paper. Perhaps modality is also an influencing factor.

[PredelinaAsya]

Thanks! I'll try to change the weights. Tell me, please, did you change any parameters in the config, except the paths, when run on EventPed?

[gitac10]

Thanks! I'll try to change the weights. Tell me, please, did you change any parameters in the config, except the paths, when run on EventPed?

i think just need simply modify the config to fit datasets,If it is modified properly, only the weight path should be modified. About modality.For example, for the same weight, in the image modality it may be "map79", and in the event modality it may become 60 or something else.perhaps you can take a closer look at the paper again and see the corresponding relationship between weights and datasets/modalities?

[iakRulan]

Thank you! I also fixed a bug with petrel_client. Now there are problems with uploading the dataset. Did you manage to run the pipeline after correcting this error?

seems i successfully run the code. i use the eventped dataset, and modified the mmpedestron_mix5datasets_best.py, get the results corresponding to the paper.i just change the dataset path,and the weight path

Hello, could you please tell me the version of Python and PyTorch you are using? I am currently experiencing some issues with my CUDA environment, like "RuntimeError: CUDA error: no kernel image is available for execution on the device" And I was wondering if your weights are being loaded from a '.pth' file in a repository. I am encountering reports of "unexpected key in source state_dict: ..." when trying to use them.

[gitac10]

Thank you! I also fixed a bug with petrel_client. Now there are problems with uploading the dataset. Did you manage to run the pipeline after correcting this error?

seems i successfully run the code. i use the eventped dataset, and modified the mmpedestron_mix5datasets_best.py, get the results corresponding to the paper.i just change the dataset path,and the weight path

Hello, could you please tell me the version of Python and PyTorch you are using? I am currently experiencing some issues with my CUDA environment, like "RuntimeError: CUDA error: no kernel image is available for execution on the device" And I was wondering if your weights are being loaded from a '.pth' file in a repository. I am encountering reports of "unexpected key in source state_dict: ..." when trying to use them.

这个issue前面讨论了python和cuda版本，torch那几个是这样

[iakRulan]

a

谢谢！我已经成功运行，请教下你有训练过模型吗？我加载了模型权重很多层都不匹配。

[iakRulan]

Thanks! I'll try to change the weights. Tell me, please, did you change any parameters in the config, except the paths, when run on EventPed?

Hello, have you trained the model and obtained your own model weights?

[PredelinaAsya]

Thanks! I'll try to change the weights. Tell me, please, did you change any parameters in the config, except the paths, when run on EventPed?

Hello, have you trained the model and obtained your own model weights?

No, I didn't train a new model, I used pre-trained weights from the repository