【学习笔记】OCR竞赛实践

概况

OCR （Optical Character Recognition，光学字符识别）是最早的计算机视觉任务之一。人类使用电子设备（图像采集装置）采集现实场景中打印的字符，然后通过检测字符的形状，用字符识别的方法将其翻译成计算机文字。

相关链接：“英特尔创新大师杯”深度学习挑战赛

环境搭建

python -m pip install paddlepaddle-gpu -i https://mirror.baidu.com/pypi/simple

baseline

主要代码如下：

model = paddle.DataParallel(model)

program.train(config, train_dataloader, valid_dataloader, device, model,
                  loss_class, optimizer, lr_scheduler, post_process_class,
                  eval_class, pre_best_model_dict, logger, vdl_writer)

def train(config,
          train_dataloader,
          valid_dataloader,
          device,
          model,
          loss_class,
          optimizer,
          lr_scheduler,
          post_process_class,
          eval_class,
          pre_best_model_dict,
          logger,
          vdl_writer=None):
......
    model.train()
......
    for epoch in range(start_epoch, epoch_num + 1):
        train_dataloader = build_dataloader(
            config, 'Train', device, logger, seed=epoch)
        
        for idx, batch in enumerate(train_dataloader):
            
            images = batch[0]
            preds = model(images)
            loss = loss_class(preds, batch)
            avg_loss = loss['loss']
            avg_loss.backward()
            optimizer.step()
            optimizer.clear_grad()
......
       save_model(
                model,
                optimizer,
                save_model_dir,
                logger,
                is_best=False,
                prefix='iter_epoch_{}'.format(epoch),
                best_model_dict=best_model_dict,
                epoch=epoch,
                global_step=global_step)     

实践结果

TBD

相关阅读

https://gitee.com/coggle/tianchi-intel-PaddleOCR