diff --git a/ppocr/data/det/dataset_traversal.py b/ppocr/data/det/dataset_traversal.py
index 5331ec191666f797d1dba50f11beb958a43b71e9..737cbe2e90dc259c1266b554ae0c6735ea3a52d2 100644
--- a/ppocr/data/det/dataset_traversal.py
+++ b/ppocr/data/det/dataset_traversal.py
@@ -96,6 +96,7 @@ class EvalTestReader(object):
                 img = cv2.imread(img_path)
                 if img is None:
                     logger.info("{} does not exist!".format(img_path))
+                    continue
                 elif len(list(img.shape)) == 2 or img.shape[2] == 1:
                     img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
                 outs = process_function(img)
diff --git a/tools/program.py b/tools/program.py
index 6a51e5c37175f0f45e87571122adc2aba04d491c..ff8743f15e2925a92fff76e7430e761d7baa720e 100755
--- a/tools/program.py
+++ b/tools/program.py
@@ -256,15 +256,15 @@ def train_eval_det_run(config, exe, train_info_dict, eval_info_dict):
                 t2 = time.time()
                 train_batch_elapse = t2 - t1
                 train_stats.update(stats)
-                if train_batch_id > start_eval_step and (train_batch_id -start_eval_step)  \
+                if train_batch_id > 0 and train_batch_id  \
                     % print_batch_step == 0:
                     logs = train_stats.log()
                     strs = 'epoch: {}, iter: {}, {}, time: {:.3f}'.format(
                         epoch, train_batch_id, logs, train_batch_elapse)
                     logger.info(strs)
 
-                if train_batch_id > 0 and\
-                    train_batch_id % eval_batch_step == 0:
+                if train_batch_id > start_eval_step and\
+                    (train_batch_id - start_eval_step) % eval_batch_step == 0:
                     metrics = eval_det_run(exe, config, eval_info_dict, "eval")
                     hmean = metrics['hmean']
                     if hmean >= best_eval_hmean: