config不合并FLAGS

Author: WenmuZhou

tools/eval.py

@@ -25,7 +25,6 @@ def main():
     cfg = Config(FLAGS.config)
     FLAGS = vars(FLAGS)
     opt = FLAGS.pop('opt')
-    cfg.merge_dict(FLAGS)
     cfg.merge_dict(opt)
     trainer = Trainer(cfg, mode='eval')
 

tools/export.py

@@ -107,6 +107,5 @@ def parse_args():
     cfg = Config(FLAGS.config)
     FLAGS = vars(FLAGS)
     opt = FLAGS.pop('opt')
-    cfg.merge_dict(FLAGS)
     cfg.merge_dict(opt)
     main(cfg, FLAGS['type'])

tools/infer_cls.py

@@ -71,6 +71,5 @@ def main(cfg):
     cfg = Config(FLAGS.config)
     FLAGS = vars(FLAGS)
     opt = FLAGS.pop('opt')
-    cfg.merge_dict(FLAGS)
     cfg.merge_dict(opt)
     main(cfg.cfg)

tools/infer_det.py

@@ -104,6 +104,5 @@ def main(cfg):
     cfg = Config(FLAGS.config)
     FLAGS = vars(FLAGS)
     opt = FLAGS.pop('opt')
-    cfg.merge_dict(FLAGS)
     cfg.merge_dict(opt)
     main(cfg.cfg)

tools/infer_rec.py

@@ -121,6 +121,5 @@ def main(cfg):
     cfg = Config(FLAGS.config)
     FLAGS = vars(FLAGS)
     opt = FLAGS.pop('opt')
-    cfg.merge_dict(FLAGS)
     cfg.merge_dict(opt)
     main(cfg.cfg)

tools/train.py

@@ -29,7 +29,6 @@ def main():
     cfg = Config(FLAGS.config)
     FLAGS = vars(FLAGS)
     opt = FLAGS.pop('opt')
-    cfg.merge_dict(FLAGS)
     cfg.merge_dict(opt)
     trainer = Trainer(cfg, mode='train_eval' if FLAGS['eval'] else 'train')
     trainer.train()

torchocr/engine/config.py

@@ -135,8 +135,6 @@ def _load_config_with_base(self, file_path):
 
             del file_cfg[self.BASE_KEY]
             file_cfg = _merge_dict(all_base_cfg, file_cfg)
-        file_cfg['filename'] = os.path.splitext(os.path.split(file_path)[-1])[
-            0]
         return file_cfg
 
     def merge_dict(self, args):

torchocr/utils/ckpt.py

@@ -54,14 +54,14 @@ def load_ckpt(model, cfg, optimizer=None, lr_scheduler=None, logger=None):
             optimizer.load_state_dict(checkpoint['optimizer'])
         if lr_scheduler is not None:
             lr_scheduler.load_state_dict(checkpoint['scheduler'])
-        logger.info(f"resume from checkpoint {checkpoints} (epoch {checkpoint['epoch']})")
+        logger.info(f"resume from checkpoint: {checkpoints} (epoch {checkpoint['epoch']})")
 
         status['global_step'] = checkpoint['global_step']
         status['epoch'] = checkpoint['epoch'] + 1
         status['metrics'] = checkpoint['metrics']
     elif pretrained_model and os.path.exists(pretrained_model):
         load_pretrained_params(model, pretrained_model)
-        logger.info(f"finetune from checkpoint {pretrained_model}")
+        logger.info(f"finetune from checkpoint: {pretrained_model}")
     else:
         logger.info("train from scratch")
     return status