Performance questions

[HoBeom]

Thank you for your good research. And thank you for revealing the code quickly. I tried to implement your code in the same environment. However, using a single GPU environment, the batch have become 1/2 size. And I got the following results. edit config line 4 : GPUS: (1,) (in my environment using 0:3080, 1:2080TI)

Posetrack 2017 val

Model	Head	Shoulder	Elbow	Wrist	Hip	Knee	Ankle	Mean
DcPose_RSN	86.3907	87.718	83.2292	76.2394	80.1681	79.1894	71.2038	80.9779

Posetrack 2018 val

Model	Head	Shoulder	Elbow	Wrist	Hip	Knee	Ankle	Mean
DcPose_RSN	83.8176	86.2703	81.4414	75.3437	77.1077	77.9727	72.2061	79.4758

The result of not achieving the performance suggested is because the batch size is small?

I would like to change the deform conv module to torchvision to run your code on CUDA11.1. https://pytorch.org/vision/stable/_modules/torchvision/ops/deform_conv.html#deform_conv2d

I also encountered an error in the posetrack 2017 test dataset.

2021-04-02 14:23:13 [engine.core.function] INFO: test: [3100/5462]      Time 1.659 (1.713)      Data 0.027s (0.083s)    Accuracy 0.000 (0.006)
2021-04-02 14:25:59 [engine.core.function] INFO: test: [3200/5462]      Time 1.659 (1.712)      Data 0.027s (0.081s)    Accuracy 0.000 (0.006)
Traceback (most recent call last):
  File "run.py", line 33, in <module>
    main()
  File "run.py", line 29, in main
    runner.launch()
  File "/DCPose/engine/defaults/runner.py", line 63, in launch
    evaluator.exec()
  File "/DCPose/engine/defaults/evaluator.py", line 20, in exec
    self.eval()
  File "/DCPose/engine/defaults/evaluator.py", line 73, in eval
    phase=self.phase)
  File "/DCPose/engine/core/function.py", line 165, in eval
    input_x, input_sup_A, input_sup_B, target_heatmaps, target_heatmaps_weight, meta = next(self.dataloader_iter)
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 345, in __next__
    data = self._next_data()
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 856, in _next_data
    return self._process_data(data)
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 881, in _process_data
    data.reraise()
  File "/opt/conda/lib/python3.7/site-packages/torch/_utils.py", line 394, in reraise
    raise self.exc_type(msg)
AttributeError: Caught AttributeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/data/_utils/worker.py", line 178, in _worker_loop
    data = fetcher.fetch(index)
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/opt/conda/lib/python3.7/site-packages/torch/utils/data/_utils/fetch.py", line 44, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/DCPose/datasets/zoo/posetrack/PoseTrack.py", line 100, in __getitem__
    return self._get_spatiotemporal_window(data_item)
  File "/DCPose/datasets/zoo/posetrack/PoseTrack.py", line 166, in _get_spatiotemporal_window
    self.logger.error(error_msg)
AttributeError: 'PoseTrack' object has no attribute 'logger'

[FRunyang]

About the performance,

1. May be some details are different.
2. We have not released our final model because it involves our next works.

Abour the error, logger is a tools to print info, you can annotate it if some errors in it.

[FRunyang]

@HoBeom

[HoBeom]

Thank you for responding. I'll figure out the error. I have changed your code to use the CUDA 11.1 version and use the deform conv module as an implementation of Torchvision (forked code in https://github.com/HoBeom/DCPose). The following results have been obtained with the same setting, and we would like to conduct a follow-up study on the basis of this study. Thank you. Posetrack 2018 val (Using torchvison.ops.deform_conv2d) batch size 32, single 2080TI 20 epoch	Model	Head	Shoulder	Elbow	Wrist	Hip	Knee	Ankle	Mean
DcPose_RSN	83.7925	86.4388	81.4788	75.82	77.8675	78.0722	72.4103	79.7035