我在同一PyCharm IDE中运行了两个几乎完全相同的脚本。他们都调用了第三个脚本，该脚本使用matplotlib将Numpy数组输出到PNG。其中一个脚本可以正常工作并输出PNG。另一个脚本引发以下错误：
qt.qpa.plugin：即使找到了Qt平台插件“ windows”，也无法在“”中加载。
脚本之间的差异很小-它们的差异仅在于它们各自导入不同的pytorch模型（均为我创建的模型）。这使我认为可能存在某种环境差异，但是它们都是在相同的IDE和Conda环境中启动的。
有人有什么建议吗？
编辑：
这是两个文件的精简版本。我删除的位在字符和空格之间都是相同的。
文件1：
import argparse
from average_meter import AverageMeter
import criteria # -> NOTE: This is not used, but was not removed in this file
import dataset
from densenet_mapmaker_model import DensenetMapMakerModel # -> Different model imported
import json
import math
import os
import time
import torch
from torch.autograd import Variable
import torch.nn as nn
from torchvision import transforms
import visualizer
parser = argparse.ArgumentParser(description='Densenet Map Maker Trainer for Steel')
# ...Define args...
##########################################################################################
# These args were defined here, but not in the other file. They are not actually used here, just defined and ignored.
parser.add_argument('--shrinkage_coef', dest='shrinkage_coef', type=int,
help='Use a shrinking gaussian loss and multiply its result by this amount.')
parser.add_argument('--multiply', dest='multiply', type=int,
help='If 1, in the shrinking gaussian loss, ' +
'multiply by 1000 before comparing and then divide the loss by 1000.')
parser.add_argument('--diff_coef', dest='diff_coef', type=int,
help='Use the count difference loss and multiply its result by this amount.')
parser.add_argument('--clustering_coef', dest='clustering_coef', type=int,
help='Use the clustering loss and multiply its result by this amount.')
##########################################################################################
use_cuda = torch.cuda.is_available()
def main():
best_mse = 1e6
version = 1
args = parser.parse_args()
# ...set up default arg values...
args.loss_calc_method = 'DenseNet' # Defined here, but not in the other file. Not used here and could be removed.
# ...load dataset lists...
model = DensenetMapMakerModel() # -> Note: Different model used
if use_cuda:
model = model.cuda()
# ...Create criteria and optimizer...
# ...Load weights from last training run...
for epoch in range(args.start_epoch, args.epochs):
train(train_list, model, criterion, optimizer, epoch, args)
mse = validate(val_list, model, criterion, epoch, args)
# ...Check result and save weights...
def train(...):
# ...Set up counters and dataset...
model.train()
end = time.time()
# img: torch.Tensor # Image data
# location : torch.Tensor. Target annotation locations for the image.
# image_id : str. Unique ID of image.
for i, (img, location, image_id) in enumerate(train_loader):
# ...Run input through model, check result and backpropagate...
if i % args.print_freq == 0:
# Note: Print statement has different whitespace formatting from other file...
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
.format(
epoch, i, len(train_loader), batch_time=batch_time,
data_time=data_time, loss=losses))
def validate(...) -> float:
# ...Set up dataset...
# img: torch.Tensor # Image data
# location : torch.Tensor. Target annotation locations for the image.
# image_id : str. Unique ID of image.
for i, (img, location, image_id) in enumerate(test_loader):
# ...Run input through model and check result...
# Output target and source the first time...
output_dir = args.output_dir
if args.output_images == 1 and epoch % 1 == 0 and i == 2: # -> NOTE: Other file does NOT have additional check i == 2
visualizer.display_tensor(output, output_dir + "/epoch_" + str(epoch) + "_" + str(i) + "_output")
if __name__ == '__main__':
main()
文件2：
import argparse
from average_meter import AverageMeter
import dataset
from steel_scanner_model import SteelScanner # -> Different model imported
import json
import math
import os
import time
import torch
from torch.autograd import Variable
import torch.nn as nn
from torchvision import transforms
import visualizer
parser = argparse.ArgumentParser(description='Densenet Map Maker Trainer for Steel')
# ...Define args...
use_cuda = torch.cuda.is_available()
def main():
best_mse = 1e6
version = 1
args = parser.parse_args()
# ...set up default arg values...
# ...load dataset lists...
model = SteelScanner() # -> Note: Different model used
if use_cuda:
model = model.cuda()
# ...Create criteria and optimizer...
# ...Load weights from last training run...
for epoch in range(args.start_epoch, args.epochs):
train(train_list, model, criterion, optimizer, epoch, args)
mse = validate(val_list, model, criterion, epoch, args)
# ...Check result and save weights...
def train(...):
# ...Set up counters and dataset...
model.train()
end = time.time()
# img: torch.Tensor # Image data
# location : torch.Tensor. Target annotation locations for the image.
# image_id : str. Unique ID of image.
for i, (img, location, image_id) in enumerate(train_loader):
# ...Run input through model, check result and backpropagate...
if i % args.print_freq == 0:
# Note: Print statement has different whitespace formatting from other file...
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
.format(epoch, i, len(train_loader), batch_time=batch_time, data_time=data_time, loss=losses))
def validate(...) -> float:
# ...Set up dataset...
# img: torch.Tensor # Image data
# location : torch.Tensor. Target annotation locations for the image.
# image_id : str. Unique ID of image.
for i, (img, location, image_id) in enumerate(test_loader):
# ...Run input through model and check result...
# Output target and source the first time...
output_dir = args.output_dir
if args.output_images == 1 and epoch % 1 == 0: # -> NOTE: Other file has additional check i == 2
# NOTE: The error occurs inside 'display_tensor'
visualizer.display_tensor(output, output_dir + "/epoch_" + str(epoch) + "_" + str(i) + "_output")
if __name__ == '__main__':
main()
从第二个文件调用时，该错误发生在visualizer.display_tensor内部。