diff --git a/PaddleCV/PaddleVideo/README.md b/PaddleCV/PaddleVideo/README.md
index 34a2b78edcff8c1cee2602d22253653e622e1c83..2b74ef5f5862ef033883ffec7ac6d7c126b3725f 100644
--- a/PaddleCV/PaddleVideo/README.md
+++ b/PaddleCV/PaddleVideo/README.md
@@ -37,6 +37,10 @@
 
 - CUDNN >= 7.0
 
+- pandas
+
+- h5py
+
 - 使用Youtube-8M数据集时，需要将tfrecord数据转化成pickle格式，需要用到Tensorflow，详见[数据说明](./data/dataset/README.md)中Youtube-8M部分。与此相关的模型是Attention Cluster, Attention LSTM, NeXtVLAD，使用其他模型请忽略此项。
 
 - 使用Kinetics数据集时，如果需要将mp4文件提前解码并保存成pickle格式，需要用到ffmpeg，详见[数据说明](./data/dataset/README.md)中Kinetics部分。需要说明的是Nonlocal模型虽然也使用Kinetics数据集，但输入数据是视频源文件，不需要提前解码，不涉及此项。与此相关的模型是TSN, TSM, StNet，使用其他模型请忽略此项。
diff --git a/PaddleCV/PaddleVideo/data/dataset/ets/README.md b/PaddleCV/PaddleVideo/data/dataset/ets/README.md
index c725f30dfc697657ed697a483c9596a418049206..a17e140a1ccfee1b5b426cbcbd8c59c77075e45c 100644
--- a/PaddleCV/PaddleVideo/data/dataset/ets/README.md
+++ b/PaddleCV/PaddleVideo/data/dataset/ets/README.md
@@ -28,7 +28,6 @@ ets
   |----feat_data/
   |----train.list
   |----val.list
-  |----preprocess.sh
   |----generate_train_pickle.py
   |----generate_data.py
   |----generate_infer_data.py
diff --git a/PaddleCV/PaddleVideo/metrics/tall_metrics/tall_metrics.py b/PaddleCV/PaddleVideo/metrics/tall_metrics/tall_metrics.py
index 0ed0f39d92d5731524a2fe559ced9f53e1d518e8..e8cee0c694830585bd5eef1264cfb4a9bde5c5a2 100644
--- a/PaddleCV/PaddleVideo/metrics/tall_metrics/tall_metrics.py
+++ b/PaddleCV/PaddleVideo/metrics/tall_metrics/tall_metrics.py
@@ -46,7 +46,7 @@ class MetricsCalculator():
         if len(x1) == 0:
             return pick
 
-        union = map(operator.sub, x2, x1)  # union = x2-x1
+        union = list(map(operator.sub, x2, x1))  # union = x2-x1
 
         I = [i[0] for i in sorted(
             enumerate(sim), key=lambda x: x[1])]  # sort and get index
diff --git a/PaddleCV/PaddleVideo/models/ets/ets.py b/PaddleCV/PaddleVideo/models/ets/ets.py
index 41a012e240421a610bb8a6f20c7017b0423b2c69..dc92fc1c5adbabd6a49f070a00d4a60e83b76924 100644
--- a/PaddleCV/PaddleVideo/models/ets/ets.py
+++ b/PaddleCV/PaddleVideo/models/ets/ets.py
@@ -189,4 +189,7 @@ class ETS(ModelBase):
         return (None, None)
 
     def weights_info(self):
-        pass
+        return (
+            'ETS_final.pdparams',
+            'https://paddlemodels.bj.bcebos.com/video_caption/ETS_final.pdparams'
+        )
diff --git a/PaddleCV/PaddleVideo/models/tall/tall.py b/PaddleCV/PaddleVideo/models/tall/tall.py
index c09eea218598412d43f1fc4b5add8eaf498b8db2..fbbdac64babd3cb85240b7578c2303aefd7cb816 100644
--- a/PaddleCV/PaddleVideo/models/tall/tall.py
+++ b/PaddleCV/PaddleVideo/models/tall/tall.py
@@ -162,4 +162,7 @@ class TALL(ModelBase):
         return (None, None)
 
     def weights_info(self):
-        pass
+        return (
+            'TALL_final.pdparams',
+            'https://paddlemodels.bj.bcebos.com/video_grounding/TALL_final.pdparams'
+        )
diff --git a/PaddleCV/PaddleVideo/predict.py b/PaddleCV/PaddleVideo/predict.py
index d21deafd5235707b8230a43d531d4429cccb5d2f..c467b8f84e40200e62096a8d4f9ef727beeb7b18 100644
--- a/PaddleCV/PaddleVideo/predict.py
+++ b/PaddleCV/PaddleVideo/predict.py
@@ -142,7 +142,7 @@ def infer(args):
         if args.model_name == 'ETS':
             data_feed_in = [items[:3] for items in data]
             vinfo = [items[3:] for items in data]
-            video_id = [items[6] for items in vinfo]
+            video_id = [items[0] for items in vinfo]
             infer_outs = exe.run(fetch_list=fetch_list,
                                  feed=infer_feeder.feed(data_feed_in),
                                  return_numpy=False)
@@ -150,7 +150,7 @@ def infer(args):
         elif args.model_name == 'TALL':
             data_feed_in = [items[:2] for items in data]
             vinfo = [items[2:] for items in data]
-            video_id = [items[0] for items in vinfo]
+            video_id = [items[6] for items in vinfo]
             infer_outs = exe.run(fetch_list=fetch_list,
                                  feed=infer_feeder.feed(data_feed_in),
                                  return_numpy=True)
diff --git a/PaddleCV/PaddleVideo/reader/tall_reader.py b/PaddleCV/PaddleVideo/reader/tall_reader.py
index b59b7718152ae21ac14d6e553ccc1695895f0175..3160561baf2fabd117ba9eaada556587940c7176 100644
--- a/PaddleCV/PaddleVideo/reader/tall_reader.py
+++ b/PaddleCV/PaddleVideo/reader/tall_reader.py
@@ -139,8 +139,8 @@ class TALLReader(DataReader):
 
         self.clip_sentence_pairs = []
         for l in cs:
-            clip_name = l[0]
-            sent_vecs = l[1]
+            clip_name = l[0].decode('utf-8')  #byte object to string
+            sent_vecs = l[1]  #numpy array
             for sent_vec in sent_vecs:
                 self.clip_sentence_pairs.append((clip_name, sent_vec))  #10146
         logger.info(self.mode.upper() + ':' + str(
@@ -185,7 +185,8 @@ class TALLReader(DataReader):
                                                       (start, end))
                                 if nIoL < 0.15:
                                     movie_length = movie_length_info[
-                                        movie_name.split(".")[0]]
+                                        movie_name.split(".")[0].encode(
+                                            'utf-8')]  #str to byte
                                     start_offset = o_start - start
                                     end_offset = o_end - end
                                     self.clip_sentence_pairs_iou.append(