用pip 安装建议用国内源，如 pip install xxx -i https://pypi.tuna.tsinghua.edu.cn/simple
目录
1.conda env 环境创建
conda create -n py39 python=3.9
2. install pytorch
先查看cuda版本 , 再对应pytorch版本
查看系统nvidia驱动版本支持最高cuda版本
查看当前cuda版本
根据对应cuda版本安装pytorch torchvision
source activate py39
conda install pytorch torchvision cudatoolkit=11.3 -c pytorch
3. install fvcore
pip install git+https://github.com/facebookresearch/fvcore
4. install simplejson
pip install simplejson
5. gcc版本查看
gcc -v
版本是 7.5.0
6. PyAV
conda install av -c conda-forge
7.ffmpeg with PyAV
pip install av
8. PyYaml , tqdm
pip list fvcore
9.iopath
pip install -U iopath
10. psutil
pip install psutil
11. opencv
pip install opencv-python
12. tensorboard
查看是否安装tensorboard:
conda list tensorboard
没有安装tensorboard
pip install tensorboard
13. moviepy
pip install moviepy
14. PyTorchVideo
pip install pytorchvideo
15. Detectron2
git clone https://github.com/facebookresearch/detectron2 detectron2_repo
pip install -e detectron2_repo
16. FairScale
pip install git+https://github.com/facebookresearch/fairscale
17. SlowFast
cd SlowFast
python setup.py build develop
运行Demo测试模型
python3 tools/run_net.py --cfg demo/AVA/SLOWFAST_32x2_R101_50_50.yaml
安装过程中遇到的一些errors
error0
not find PIL
解决办法：将setup.py 中的 PIL 更改为 Pillow
error1
from pytorchvideo.layers.distributed import ( # noqa
ImportError: cannot import name 'cat_all_gather' from 'pytorchvideo.layers.distributed' (/home/cxgk/anaconda3/envs/sf/lib/python3.9/site-packages/pytorchvideo/layers/distributed.py)
解决方式：
方式一：将pytorchvideo/pytorchvideo at main · facebookresearch/pytorchvideo · GitHub文件下内容复制到虚拟环境所对应的文件下，这里是：/home/cxgk/anaconda3/envs/sf/lib/python3.9/site-packages/pytorchvideo/
方式二：
layers/distributed.py添加如下内容
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
"""Distributed helpers."""
import torch
import torch.distributed as dist
from torch._C._distributed_c10d import ProcessGroup
from torch.autograd.function import Function
_LOCAL_PROCESS_GROUP = None
def get_world_size() -> int:
"""
Simple wrapper for correctly getting worldsize in both distributed
/ non-distributed settings
"""
return (
torch.distributed.get_world_size()
if torch.distributed.is_available() and torch.distributed.is_initialized()
else 1
)
def cat_all_gather(tensors, local=False):
"""Performs the concatenated all_reduce operation on the provided tensors."""
if local:
gather_sz = get_local_size()
else:
gather_sz = torch.distributed.get_world_size()
tensors_gather = [torch.ones_like(tensors) for _ in range(gather_sz)]
torch.distributed.all_gather(
tensors_gather,
tensors,
async_op=False,
group=_LOCAL_PROCESS_GROUP if local else None,
)
output = torch.cat(tensors_gather, dim=0)
return output
def init_distributed_training(cfg):
"""
Initialize variables needed for distributed training.
"""
if cfg.NUM_GPUS <= 1:
return
num_gpus_per_machine = cfg.NUM_GPUS
num_machines = dist.get_world_size() // num_gpus_per_machine
for i in range(num_machines):
ranks_on_i = list(
range(i * num_gpus_per_machine, (i + 1) * num_gpus_per_machine)
)
pg = dist.new_group(ranks_on_i)
if i == cfg.SHARD_ID:
global _LOCAL_PROCESS_GROUP
_LOCAL_PROCESS_GROUP = pg
def get_local_size() -> int:
"""
Returns:
The size of the per-machine process group,
i.e. the number of processes per machine.
"""
if not dist.is_available():
return 1
if not dist.is_initialized():
return 1
return dist.get_world_size(group=_LOCAL_PROCESS_GROUP)
def get_local_rank() -> int:
"""
Returns:
The rank of the current process within the local (per-machine) process group.
"""
if not dist.is_available():
return 0
if not dist.is_initialized():
return 0
assert _LOCAL_PROCESS_GROUP is not None
return dist.get_rank(group=_LOCAL_PROCESS_GROUP)
def get_local_process_group() -> ProcessGroup:
assert _LOCAL_PROCESS_GROUP is not None
return _LOCAL_PROCESS_GROUP
class GroupGather(Function):
"""
GroupGather performs all gather on each of the local process/ GPU groups.
"""
@staticmethod
def forward(ctx, input, num_sync_devices, num_groups):
"""
Perform forwarding, gathering the stats across different process/ GPU
group.
"""
ctx.num_sync_devices = num_sync_devices
ctx.num_groups = num_groups
input_list = [torch.zeros_like(input) for k in range(get_local_size())]
dist.all_gather(
input_list, input, async_op=False, group=get_local_process_group()
)
inputs = torch.stack(input_list, dim=0)
if num_groups > 1:
rank = get_local_rank()
group_idx = rank // num_sync_devices
inputs = inputs[
group_idx * num_sync_devices : (group_idx + 1) * num_sync_devices
]
inputs = torch.sum(inputs, dim=0)
return inputs
@staticmethod
def backward(ctx, grad_output):
"""
Perform backwarding, gathering the gradients across different process/ GPU
group.
"""
grad_output_list = [
torch.zeros_like(grad_output) for k in range(get_local_size())
]
dist.all_gather(
grad_output_list,
grad_output,
async_op=False,
group=get_local_process_group(),
)
grads = torch.stack(grad_output_list, dim=0)
if ctx.num_groups > 1:
rank = get_local_rank()
group_idx = rank // ctx.num_sync_devices
grads = grads[
group_idx
* ctx.num_sync_devices : (group_idx + 1)
* ctx.num_sync_devices
]
grads = torch.sum(grads, dim=0)
return grads, None, Noneerror2
from scipy.ndimage import gaussian_filter
ModuleNotFoundError: No module named 'scipy'
解决方法：
pip install scipy
error3
from av._core import time_base, library_versions
ImportError: /home/cxgk/anaconda3/envs/sf/lib/python3.9/site-packages/av/../../.././libgnutls.so.30: symbol mpn_copyi version HOGWEED_6 not defined in file libhogweed.so.6 with link time reference
解决方法：
先移处av包
使用 pip安装
pip install av
error4
File "/media/cxgk/Linux/work/SlowFast/slowfast/models/losses.py", line 11, in
from pytorchvideo.losses.soft_target_cross_entropy import (
ModuleNotFoundError: No module named 'pytorchvideo.losses'
解决办法：
打开"/home/cxgk/anaconda3/envs/sf/lib/python3.9/site-packages/pytorchvideo/losses"，在文件夹下新建 soft_target_cross_entropy.py， 并打开添加如下代码：
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
import torch
import torch.nn as nn
import torch.nn.functional as F
from pytorchvideo.layers.utils import set_attributes
from pytorchvideo.transforms.functional import convert_to_one_hot
class SoftTargetCrossEntropyLoss(nn.Module):
"""
Adapted from Classy Vision: ./classy_vision/losses/soft_target_cross_entropy_loss.py.
This allows the targets for the cross entropy loss to be multi-label.
"""
def __init__(
self,
ignore_index: int = -100,
reduction: str = "mean",
normalize_targets: bool = True,
) -> None:
"""
Args:
ignore_index (int): sample should be ignored for loss if the class is this value.
reduction (str): specifies reduction to apply to the output.
normalize_targets (bool): whether the targets should be normalized to a sum of 1
based on the total count of positive targets for a given sample.
"""
super().__init__()
set_attributes(self, locals())
assert isinstance(self.normalize_targets, bool)
if self.reduction not in ["mean", "none"]:
raise NotImplementedError(
'reduction type "{}" not implemented'.format(self.reduction)
)
self.eps = torch.finfo(torch.float32).eps
def forward(self, input: torch.Tensor, target: torch.Tensor) -> torch.Tensor:
"""
Args:
input (torch.Tensor): the shape of the tensor is N x C, where N is the number of
samples and C is the number of classes. The tensor is raw input without
softmax/sigmoid.
target (torch.Tensor): the shape of the tensor is N x C or N. If the shape is N, we
will convert the target to one hot vectors.
"""
# Check if targets are inputted as class integers
if target.ndim == 1:
assert (
input.shape[0] == target.shape[0]
), "SoftTargetCrossEntropyLoss requires input and target to have same batch size!"
target = convert_to_one_hot(target.view(-1, 1), input.shape[1])
assert input.shape == target.shape, (
"SoftTargetCrossEntropyLoss requires input and target to be same "
f"shape: {input.shape} != {target.shape}"
)
# Samples where the targets are ignore_index do not contribute to the loss
N, C = target.shape
valid_mask = torch.ones((N, 1), dtype=torch.float).to(input.device)
if 0 <= self.ignore_index <= C - 1:
drop_idx = target[:, self.ignore_idx] > 0
valid_mask[drop_idx] = 0
valid_targets = target.float() * valid_mask
if self.normalize_targets:
valid_targets /= self.eps + valid_targets.sum(dim=1, keepdim=True)
per_sample_per_target_loss = -valid_targets * F.log_softmax(input, -1)
per_sample_loss = torch.sum(per_sample_per_target_loss, -1)
# Perform reduction
if self.reduction == "mean":
# Normalize based on the number of samples with > 0 non-ignored targets
loss = per_sample_loss.sum() / torch.sum(
(torch.sum(valid_mask, -1) > 0)
).clamp(min=1)
elif self.reduction == "none":
loss = per_sample_loss
return error5
from sklearn.metrics import confusion_matrix
ModuleNotFoundError: No module named 'sklearn'
解决办法：
pip install scikit-learn
error6
raise KeyError("Non-existent config key: {}".format(full_key))
KeyError: 'Non-existent config key: TENSORBOARD.MODEL_VIS.TOPK'
解决方法：
注释掉如下三行：
TENSORBOARD
MODEL_VIS
TOPK
error7
RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 3.94 GiB total capacity; 2.83 GiB already allocated; 25.44 MiB free; 2.84 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
解决方法：
将yaml里的帧数改小：
DATA:
NUM_FRAMES: 16
Reference:
https://github.com/facebookresearch/pytorchvideo/blob/main/pytorchvideo