解决pytorch版PseNet源码报错 ：UnboundLocalError: local variable 'score_text' referenced before assignment

最近在折腾Psenet模型，实现的过程有些艰辛。

原作者的代码基于python2.7，我自己只有python3环境，一路解决python不同版本的问题，碰到一个大BUG！

eckpoint path: checkpoints/ic15_resnet50_bs_2_ep_600
init lr: 0.00100000
('schedule: ', [200, 400])
Training from scratch.

Epoch: [1 | 600] LR: 0.001000
Traceback (most recent call last):
File "train_ic15.py", line 289, in 
main(args)
File "train_ic15.py", line 257, in main
train_loss, train_te_acc, train_ke_acc, train_te_iou, train_ke_iou = train(train_loader, model, dice_loss, optimizer, epoch)
File "train_ic15.py", line 175, in train
return (losses.avg, score_text['Mean Acc'], score_kernel['Mean Acc'], IC15Loaderscore_text['Mean IoU'], score_kernel['Mean IoU'])
UnboundLocalError: local variable 'score_text' referenced before assignment

附图：

说是局部变量 score_text被重复引用，一路调试发现还是python3和python2的坑，整除操作运算符不同！！！

icdar2015_loader.py 的198行，这样修改：

bboxes = np.reshape(bboxes * ([img.shape[1], img.shape[0]] * 4), (bboxes.shape[0], bboxes.shape[1] // 2, 2)).astype('int32') # py3

# bboxes = np.reshape(bboxes * ([img.shape[1], img.shape[0]] * 4), (bboxes.shape[0], bboxes.shape[1] / 2, 2)).astype('int32') # py2

 

这样就可以正常运行啦！