def val(net, dataset, criterion, max_iter=100):
print('Start val')
for p in net.parameters():
p.requires_grad = False
num_correct, num_all = val_model(config.val_infofile,
net,
True,
log_file='compare-' +
config.saved_model_prefix + '.log')
accuracy = num_correct / num_all
print('ocr_acc: %f' % (accuracy))
if config.use_log:
with open(log_filename, 'a') as f:
f.write('ocr_acc:{}\n'.format(accuracy))
global best_acc
if accuracy > best_acc:
best_acc = accuracy
torch.save(
crnn.state_dict(),
'{}/{}_{}_{}.pth'.format(config.saved_model_dir,
config.saved_model_prefix, epoch,
int(best_acc * 1000)))
torch.save(
crnn.state_dict(), '{}/{}.pth'.format(config.saved_model_dir,
config.saved_model_prefix))
def main(crnn, train_loader, val_loader, criterion, optimizer):
crnn = crnn.to(device)
criterion = criterion.to(device)
for i, epoch in enumerate(range(params.epochs)):
# if i<1:
train(crnn, train_loader, criterion, epoch)
# # ## max_i: cut down the consuming time of testing, if you'd like to validate on the whole testset, please set it to len(val_loader)
accuracy = val(crnn, val_loader, criterion, epoch, max_i=1000)
for p in crnn.parameters():
p.requires_grad = True
# if accuracy > params.best_accuracy:
torch.save(
crnn.state_dict(),
'{0}/crnn_Rec_done_{1}_{2}.pth'.format(params.experiment, epoch,
accuracy))
torch.save(crnn.state_dict(),
'{0}/crnn_best.pth'.format(params.experiment))
print("is best accuracy: {0}".format(accuracy > params.best_accuracy))
preds = preds.view(-1, preds_size.item(), 1, 23)
for idx, head in enumerate(preds):
_, preds1 = head.max(2)
preds1 = preds1.transpose(1, 0).contiguous().view(-1)
raw_pred = converter.decode(preds1.data,
preds_size.data,
raw=False)
postpro = re.findall(temp, raw_pred)
sim_pred = postpro[0] if postpro != [] else 'Unknown'
result[idx] = sim_pred
if result[0] == 'Unknown':
fp_1 += 1
fp_2 += 1
elif result[0] == y[0]:
tp_1 += 1
if result[0] == result[1]:
tp_2 += 1
else:
fp_2 += 1
elif result[0] != y[0]:
fn_1 += 1
if result[0] == result[1]:
fn_2 += 1
torch.save(crnn.state_dict(), '/content/drive/My Drive/WeightNet/OCR(1.6)')
# numLoss = 0
# else:
cost = trainBatch(crnn, criterion, optimizer)
loss_avg.add(cost)
i += 1
# if i % opt.displayInterval == 0:
# print('[%d/%d][%d/%d] Loss: %f' % (epoch, opt.niter, i, len(train_loader), loss_avg.val()))
# loss_avg.reset()
if i % opt.valInterval == 0:
testLoss, accuracy = val(crnn, criterion)
# print('Test loss: %f, accuray: %f' % (testLoss, accuracy))
print("epoch: {}, step: {}, train loss: {:.3f}, test loss: {:.3f}, test accuracy: {:.3f}".format(epoch, num, loss_avg.val(), testLoss, accuracy))
loss_avg.reset()
num += 1
# lasttestLoss = min(lasttestLoss, testLoss)
# checkpoint
if lasttestLoss > testLoss:
print("Get a best val loss, model saving...\n")
lasttestLoss = testLoss
# delete(opt.experiment) # 删除历史模型
torch.save(crnn.state_dict(), '{}/netCRNN.pth'.format(opt.experiment))
numLoss = 0
else:
numLoss += 1
adjust_learning_rate(optimizer)
trainloader, validloader = prepare_dataloader()
crnn = crnn.CRNN(num_classes).to(device)
criterion = torch.nn.CTCLoss().to(device)
if args.adam:
optimizer = optim.Adam(crnn.parameters(), lr=args.lr)
elif args.rmsprop:
optimizer = optim.RMSprop(crnn.parameters(), lr=args.lr)
else:
optimizer = optim.Adadelta(crnn.parameters())
if args.pretrained != '':
print('loading pretrained model from {}'.format(args.pretrained))
crnn.load_state_dict(torch.load(args.pretrained))
crnn.train()
for epoch in range(args.num_epoch):
train(trainloader, crnn, converter, criterion, optimizer)
if epoch % args.eval_epoch == 0:
print('-------------------- eval --------------------')
crnn.eval()
validate(validloader, crnn, converter)
crnn.train()
if epoch % args.save_epoch == 0:
torch.save(crnn.state_dict(),
'{}/crnn_{}.pth'.format(args.save_path, epoch + 1))