Problem about input_size for customized dataset #34

KamiCalcium · 2021-06-15T19:07:19Z

Hi thanks for the great works!

I've been trying testing the pre-trained model on cityperson dataset under the cityscape dataset.

The code perfectly went through when the input_size was setting as (256, 256) for Human3.6M as default. But obviously, the output root location was not correct due to the scaling problem. The proper input_size of the cityperson image is (2048, 1024). However, if I set the size to (2048, 1024) I got the following error:

06-15 11:35:20 Creating dataset...
loading annotations into memory...
Done (t=0.04s)
creating index...
index created!
06-15 11:35:21 Load checkpoint from /dump/algopre/c-szan/github/3DMPPE_ROOTNET_RELEASE/main/../output/model_dump/snapshot_18.pth.tar
06-15 11:35:21 Creating graph...
  0%|                                                                                                              | 0/33 [00:00<?, ?it/s]THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1607370193460/work/aten/src/THC/THCCachingHostAllocator.cpp line=278 error=700 : an illegal memory access was encountered
  0%|                                                                                                              | 0/33 [02:02<?, ?it/s]
Traceback (most recent call last):
  File "test.py", line 54, in <module>
    main()
  File "test.py", line 45, in main
    coord_out = tester.model(input_img, cam_param)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 161, in forward
    outputs = self.parallel_apply(replicas, inputs, kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 171, in parallel_apply
    return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 86, in parallel_apply
    output.reraise()
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/_utils.py", line 428, in reraise
    raise self.exc_type(msg)
RuntimeError: Caught RuntimeError in replica 0 on device 0.
Original Traceback (most recent call last):
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 61, in _worker
    output = module(*input, **kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/github/3DMPPE_ROOTNET_RELEASE/main/model.py", line 100, in forward
    fm = self.backbone(input_img)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/github/3DMPPE_ROOTNET_RELEASE/main/../common/nets/resnet.py", line 64, in forward
    x = self.layer4(x)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/container.py", line 117, in forward
    input = module(input)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torchvision/models/resnet.py", line 116, in forward
    identity = self.downsample(x)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/container.py", line 117, in forward
    input = module(input)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 423, in forward
    return self._conv_forward(input, self.weight)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 420, in _conv_forward
    self.padding, self.dilation, self.groups)
RuntimeError: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR
You can try to repro this exception using the following code snippet. If that doesn't trigger the error, please include your original repro script when reporting this issue.

import torch
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.benchmark = True
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.allow_tf32 = True
data = torch.randn([32, 1024, 128, 64], dtype=torch.float, device='cuda', requires_grad=True)
net = torch.nn.Conv2d(1024, 2048, kernel_size=[1, 1], padding=[0, 0], stride=[2, 2], dilation=[1, 1], groups=1)
net = net.cuda().float()
out = net(data)
out.backward(torch.randn_like(out))
torch.cuda.synchronize()

ConvolutionParams
    data_type = CUDNN_DATA_FLOAT
    padding = [0, 0, 0]
    stride = [2, 2, 0]
    dilation = [1, 1, 0]
    groups = 1
    deterministic = false
    allow_tf32 = true
input: TensorDescriptor 0x7f260a187ca0
    type = CUDNN_DATA_FLOAT
    nbDims = 4
    dimA = 32, 1024, 128, 64,
    strideA = 8388608, 8192, 64, 1,
output: TensorDescriptor 0x7f2ed40a1160
    type = CUDNN_DATA_FLOAT
    nbDims = 4
    dimA = 32, 2048, 64, 32,
    strideA = 4194304, 2048, 32, 1,
weight: FilterDescriptor 0x7f2ed40a2b40
    type = CUDNN_DATA_FLOAT
    tensor_format = CUDNN_TENSOR_NCHW
    nbDims = 4
    dimA = 2048, 1024, 1, 1,
Pointer addresses:
    input: 0x7f1abc000000
    output: 0x7f1a7c000000
    weight: 0x7f3006a00000


terminate called after throwing an instance of 'c10::Error'
  what():  CUDA error: an illegal memory access was encountered
Exception raised from create_event_internal at /opt/conda/conda-bld/pytorch_1607370193460/work/c10/cuda/CUDACachingAllocator.cpp:687 (most recent call first):
frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x42 (0x7f305a5d18b2 in /dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/lib/libc10.so)
frame #1: c10::cuda::CUDACachingAllocator::raw_delete(void*) + 0xad2 (0x7f305a823982 in /dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/lib/libc10_cuda.so)
frame #2: c10::TensorImpl::release_resources() + 0x4d (0x7f305a5bcb7d in /dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/lib/libc10.so)
frame #3: <unknown function> + 0x5fe1ea (0x7f309bd8e1ea in /dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/lib/libtorch_python.so)
frame #4: <unknown function> + 0x5fe296 (0x7f309bd8e296 in /dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/lib/libtorch_python.so)
<omitting python frames>
frame #19: __libc_start_main + 0xf0 (0x7f30cd22f840 in /lib/x86_64-linux-gnu/libc.so.6)

Aborted (core dumped)

To debug, I then make the input_size to (1024,1024). Then it perfectly went through and the result makes more sense than when I set it to (256, 256). I also tried (2048, 2048) and it gave a different error:

06-15 11:45:01 Creating dataset...
loading annotations into memory...
Done (t=0.02s)
creating index...
index created!
06-15 11:45:02 Load checkpoint from /dump/algopre/c-szan/github/3DMPPE_ROOTNET_RELEASE/main/../output/model_dump/snapshot_18.pth.tar
06-15 11:45:02 Creating graph...
  0%|                                                                                                              | 0/33 [02:14<?, ?it/s]
Traceback (most recent call last):
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 872, in _try_get_data
    data = self._data_queue.get(timeout=timeout)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/queue.py", line 173, in get
    self.not_empty.wait(remaining)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/threading.py", line 299, in wait
    gotit = waiter.acquire(True, timeout)
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/utils/data/_utils/signal_handling.py", line 66, in handler
    _error_if_any_worker_fails()
RuntimeError: DataLoader worker (pid 85820) is killed by signal: Killed.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "test.py", line 54, in <module>
    main()
  File "test.py", line 43, in main
    for itr, (input_img, cam_param) in enumerate(tqdm(tester.batch_generator)):
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/tqdm/std.py", line 1178, in __iter__
    for obj in iterable:
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 435, in __next__
    data = self._next_data()
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 1068, in _next_data
    idx, data = self._get_data()
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 1024, in _get_data
    success, data = self._try_get_data()
  File "/dump/algopre/c-szan/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 885, in _try_get_data
    raise RuntimeError('DataLoader worker (pid(s) {}) exited unexpectedly'.format(pids_str)) from e
RuntimeError: DataLoader worker (pid(s) 85820) exited unexpectedly

I don't know what is going on when the input_size is different. Can you please look at the problem and offer me some clues?

Thanks a lot!

The text was updated successfully, but these errors were encountered:

mks0601 · 2021-06-16T04:26:56Z

Is the input image contain a single person? This algorithm requires a single person input. And I guess 1024 and 2048 is too large and the errors are about CPU/GPU memory explosion.

KamiCalcium · 2021-06-16T17:35:30Z

Is the input image contain a single person? This algorithm requires a single person input. And I guess 1024 and 2048 is too large and the errors are about CPU/GPU memory explosion.

Hey thanks for the quick response!~

I don't understand why the algorithm requires a single-person input. In your demo (people skiing), there are also multiple people, right?

About the CPU/GPU memory explosion, I'm using 4 RTX 8000. If that's still not sufficient enough, is there a way to properly train and test the image with (2048,1024)? Should I scale the input image or what will you suggest?

Thanks a lot!

mks0601 · 2021-06-16T23:23:13Z

It crops and processes each human separately.
Please read this line

KamiCalcium · 2021-06-17T21:09:15Z

It crops and processes each human separately.
Please read this line

Okay, this I understand. But what about this?

3DMPPE_ROOTNET_RELEASE/main/config.py

Line 34 in a199d50

input_shape = (256, 256)

What is this input_size then? Isn't it the whole image size? Or the maximum size for each bbox?

mks0601 · 2021-06-17T23:22:05Z

Please read the instructions carefully. here.
It also crops and processes each human separately.

KamiCalcium · 2021-06-18T16:57:47Z

Please read the instructions carefully. here.
It also crops and processes each human separately.

I read it again but I only see this description about the input_size for config.py.

In the main/config.py, you can change settings of the model including dataset to use, network backbone, and input size and so on.

My understanding is this is the whole image size right? Since human36M' size is (256,256), which is the same as the default one in config.py. Then if my picture input_size is (2048,1024), should I change it to (2048,1024)?

mks0601 · 2021-06-18T18:08:21Z

The instruction of Quick demo says Set bbox_list at here.

KamiCalcium · 2021-06-18T18:56:31Z

The instruction of Quick demo says Set bbox_list at here.

OK, this is pretty clear to me and I already set the individual bbox in the bbox_list.

But in

3DMPPE_ROOTNET_RELEASE/demo/demo.py

Line 97 in a199d50

vis_root[0] = root_3d[0] / cfg.output_shape[1] * cfg.input_shape[1]

demo.py references config.py and uses its input_shape to derive the vis_root array. I want to know what input_shape does here. Is this the whole image size?

mks0601 · 2021-06-19T00:27:52Z

No. The code crops and resizes each person to input_img_shape.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Problem about input_size for customized dataset #34

Problem about input_size for customized dataset #34

KamiCalcium commented Jun 15, 2021

mks0601 commented Jun 16, 2021

KamiCalcium commented Jun 16, 2021

mks0601 commented Jun 16, 2021

KamiCalcium commented Jun 17, 2021

mks0601 commented Jun 17, 2021

KamiCalcium commented Jun 18, 2021

mks0601 commented Jun 18, 2021

KamiCalcium commented Jun 18, 2021 •

edited

Loading

mks0601 commented Jun 19, 2021

Problem about input_size for customized dataset #34

Problem about input_size for customized dataset #34

Comments

KamiCalcium commented Jun 15, 2021

mks0601 commented Jun 16, 2021

KamiCalcium commented Jun 16, 2021

mks0601 commented Jun 16, 2021

KamiCalcium commented Jun 17, 2021

mks0601 commented Jun 17, 2021

KamiCalcium commented Jun 18, 2021

mks0601 commented Jun 18, 2021

KamiCalcium commented Jun 18, 2021 • edited Loading

mks0601 commented Jun 19, 2021

KamiCalcium commented Jun 18, 2021 •

edited

Loading