diff --git a/src/features/build_tfrecords.py b/src/features/build_tfrecords.py
new file mode 100644
index 0000000..1320e68
--- /dev/null
+++ b/src/features/build_tfrecords.py
@@ -0,0 +1,70 @@
+# -*- coding: utf-8 -*-
+import click
+import logging
+import os
+import sys
+import pandas as pd
+import tensorflow as tf
+import numpy as np
+from pathlib import Path
+from dotenv import find_dotenv, load_dotenv
+from PIL import Image
+
+
+load_dotenv(find_dotenv())
+
+def main():
+    """ Create train and test test
+    """
+
+    logger.info('Create train & test set')
+
+    # create refs labels
+    refs = pd.read_csv(refs_path)
+    refs = refs.pivot_table(index = "image", columns = "label", aggfunc = len, fill_value = 0).reset_index()
+    refs.rename(columns = lambda x: "label" + str(x) if x != 'image' else x, inplace = True)
+    refs = refs[refs['image'].isin(imgs_list)]
+
+    # initiate tf file writer
+    writer = tf.python_io.TFRecordWriter(train_filename)
+
+    for i, row in refs.iterrows():
+        img_full_path = os.path.join(imgs_path, row['image'])
+        path, raw  = loadImg(img_full_path)
+
+        example = tf.train.Example(features = tf.train.Features(feature = {
+            'input' : _bytes_feature(raw),
+            'label' : _int64_feature(row['label1'])
+        }))
+
+        logger.info("Creation TFRECORDS {}/{} : {}".format(i+1, refs.index.size, img_full_path))
+        writer.write(example.SerializeToString())
+
+    writer.close()
+
+
+def _int64_feature(value):
+    return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))
+
+def _bytes_feature(value):
+    return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value]))
+
+def loadImg(path):
+
+    raw = np.asarray(Image.open(path)).tostring()
+    
+    return path, raw
+
+
+if __name__ == '__main__':
+    log_fmt = '%(asctime)s - %(name)s - %(levelname)s - %(message)s'
+    logging.basicConfig(level=logging.INFO, format=log_fmt)
+
+    project_dir = Path(__file__).resolve().parents[2]
+    logger = logging.getLogger(__name__)
+    imgs_path = os.path.join(str(project_dir), 'data', 'interim', 'collections')
+    imgs_list = os.listdir(imgs_path)
+    refs_path = os.path.join(str(project_dir), 'data', 'external', 'refs', 'references_labels.csv')
+    train_filename = os.path.join(str(project_dir), 'data', 'processed', 'train.tfrecords')   
+
+    main()