diff --git a/README.md b/README.md
index ad49919..16b6918 100644
--- a/README.md
+++ b/README.md
@@ -1,2 +1,112 @@
-# frozen-in-time
+# Video-Text Retrieval Embedding with Frozen In Time
+
+*author: Jinling Xu*
+
+
+<br />
+
+
+
+## Description
+
+This operator extracts features for video or text with [Frozen In Time](https://arxiv.org/abs/2104.00650) which can generate embeddings for text and video by jointly training a video encoder and text encoder to maximize the cosine similarity.
+
+
+<br />
+
+
+## Code Example
+
+Load a video from path './demo_video.mp4' to generate a video embedding. 
+
+Read the text 'kids feeding and playing with the horse' to generate a text embedding. 
+
+ *Write the pipeline in simplified style*:
+
+- Encode video (default):
+```python
+import towhee
+towhee.dc(['./demo_video.mp4']) \
+        .video_decode.ffmpeg(sample_type='uniform_temporal_subsample', args={'num_samples': 4}) \
+        .runas_op(func=lambda x: [y for y in x]) \
+        .video_text_embedding.frozen_in_time(model_name='frozen_in_time_base_16_244', modality='video', device='cpu') \
+        .show()
+
+```
+- Encode text:
+```python
+import towhee
+
+towhee.dc(['kids feeding and playing with the horse']) \
+      .video_text_embedding.frozen_in_time(model_name='frozen_in_time_base_16_244', modality='text', device='cpu') \
+      .show()
+```
+
+*Write a same pipeline with explicit inputs/outputs name specifications:*
+
+```python
+import towhee
+
+towhee.dc['path'](['./demo_video.mp4']) \
+        .video_decode.ffmpeg['path', 'frames'](sample_type='uniform_temporal_subsample', args={'num_samples': 4}) \
+        .runas_op['frames', 'frames'](func=lambda x: [y for y in x]) \
+        .video_text_embedding.frozen_in_time['frames', 'vec'](model_name='frozen_in_time_base_16_244', modality='video', device='cpu') \
+        .show()
+
+towhee.dc['text'](["kids feeding and playing with the horse"]) \
+      .video_text_embedding.frozen_in_time['text','vec'](model_name='frozen_in_time_base_16_244', modality='text', device='cpu') \
+      .select['text', 'vec']() \
+      .show()
+```
+
+
+<br />
+
+
+
+## Factory Constructor
+
+Create the operator via the following factory method
+
+***frozen_in_time(model_name, modality, weight_path)***
+
+**Parameters:**
+
+​   ***model_name:*** *str*
+
+​   The model name of frozen in time. Supported model names: 
+- frozen_in_time_base_16_244
+
+
+​   ***modality:*** *str*
+
+​   Which modality(*video* or *text*) is used to generate the embedding. 
+
+​   ***weight_path:*** *str*
+
+​   pretrained model weights path.  
+
+<br />
+
+
+
+## Interface
+
+An video-text embedding operator takes a list of [Towhee VideoFrame](link/to/towhee/image/api/doc) or string as input and generate an embedding in ndarray.
+
+
+**Parameters:**
+
+​	***data:*** *List[towhee.types.Image]*  or *str*
+
+​  The data (list of  Towhee VideoFrame (which is uniform subsampled from a video) or text based on specified modality) to generate embedding.	
+
+
+
+**Returns:** *numpy.ndarray*
+
+​   The data embedding extracted by model.
+
+
+
 
diff --git a/__init__.py b/__init__.py
new file mode 100644
index 0000000..9f987c8
--- /dev/null
+++ b/__init__.py
@@ -0,0 +1,20 @@
+# Copyright 2021 Zilliz. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .frozen_in_time import FrozenInTime
+
+
+def frozen_in_time(model_name: str, modality: str, **kwargs):
+    return FrozenInTime(model_name, modality, **kwargs)
+
diff --git a/demo_video.mp4 b/demo_video.mp4
new file mode 100755
index 0000000..e6fb645
Binary files /dev/null and b/demo_video.mp4 differ
diff --git a/frozen_in_time.py b/frozen_in_time.py
new file mode 100644
index 0000000..4aaece4
--- /dev/null
+++ b/frozen_in_time.py
@@ -0,0 +1,112 @@
+# Copyright 2021 Zilliz. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy
+import numpy as np
+import torch
+
+from typing import List, Union
+from torchvision import transforms
+from towhee.models import frozen_in_time
+from towhee.operator.base import NNOperator
+from towhee import register
+from PIL import Image as PILImage
+from towhee.types import VideoFrame
+from towhee.models.utils.video_transforms import transform_video, get_configs
+from pathlib import Path
+from transformers import AutoTokenizer
+
+
+@register(output_schema=['vec'])
+class FrozenInTime(NNOperator):
+    """
+    extracts features for video or text with Frozen In Time model
+    Args:
+        model_name (str):
+            Frozen In Time model name to be used in FrozenInTime
+        modality (str):
+            Flag to decide what to return
+                - 'video': return video embedding
+                - 'text': return a dense of text embeddings
+        weight_path (str):
+            Pretrained model weights
+        device (str):
+            the device to run model
+    """
+
+    def __init__(self, model_name: str = 'frozen_in_time_base_16_244', modality: str = 'video',
+                 weight_path: str = None,
+                 device: str = None):
+        super().__init__()
+        self.model_name = model_name
+        self.modality = modality
+        if weight_path is None:
+            weight_path = str(Path(__file__).parent / 'frozen_in_time_base_16_224.pth')
+        if device is None:
+            self.device = "cuda" if torch.cuda.is_available() else "cpu"
+        else:
+            self.device = device
+        self.num_frames = 4
+        self.model = frozen_in_time.FrozenInTime(img_size=224,
+                                                 patch_size=16,
+                                                 in_chans=3,
+                                                 num_frames=self.num_frames,
+                                                 attention_style='frozen_in_time',
+                                                 is_pretrained=True,
+                                                 weights_path=weight_path,
+                                                 projection_dim=256,
+                                                 video_pretrained_model='vit_base_16x224',
+                                                 video_is_load_pretrained=False,
+                                                 video_model_type='SpaceTimeTransformer',
+                                                 text_is_load_pretrained=False,
+                                                 device=device)
+
+        self.tokenizer = AutoTokenizer.from_pretrained('distilbert-base-uncased', TOKENIZERS_PARALLELISM=False)
+        self.transform_cfgs = get_configs(
+            side_size=224,
+            crop_size=224,
+            num_frames=self.num_frames,
+            mean=[0.48145466, 0.4578275, 0.40821073],
+            std=[0.26862954, 0.26130258, 0.27577711],
+        )
+        self.model.eval()
+
+    def __call__(self, data: Union[List[VideoFrame], List[str]]):
+        if self.modality == 'video':
+            vec = self._inference_from_video(data)
+        elif self.modality == 'text':
+            vec = self._inference_from_text(data)
+        else:
+            raise ValueError("modality[{}] not implemented.".format(self._modality))
+        return vec
+
+    def _inference_from_text(self, text: List[str]):
+        text_data = self.tokenizer(text, return_tensors='pt')
+        # text_data = torch.tensor(text)
+        text_data = text_data.to(self.device)
+        text_features = self.model.compute_text(text_data)
+        return text_features.squeeze(0).detach().flatten().cpu().numpy()
+
+    def _inference_from_video(self, data: List[VideoFrame]):
+        # Convert list of towhee.types.Image to numpy.ndarray in float32
+        video = numpy.stack([img.astype(numpy.float32) / 255. for img in data], axis=0)
+        assert len(video.shape) == 4
+        video = video.transpose(3, 0, 1, 2)  # twhc -> ctwh
+        video = transform_video(
+            video=video,
+            **self.transform_cfgs
+        )
+        video = video.to(self.device)[None, ...].transpose(1, 2)
+        visual_features = self.model.compute_video(video)
+        return visual_features.squeeze(0).detach().flatten().cpu().numpy()
+
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..aa91a4e
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,3 @@
+transformers>=4.19.2
+einops>=0.4.1
+timm>=0.4.12
\ No newline at end of file