diff --git a/ACL_TensorFlow/contrib/cv/Slot-Attention_ID2028_for_ACL/README.md b/ACL_TensorFlow/contrib/cv/Slot-Attention_ID2028_for_ACL/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..2b7511abe81e13ba6ae2441ab545787d13da98ac
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/Slot-Attention_ID2028_for_ACL/README.md
@@ -0,0 +1,65 @@
+<h2 id="概述.md">概述</h2>
+提出了Slot Attention 模块，建立了感知表征 (perceptual representations, 如CNN 输出) 与 slots 之间的桥梁 (Feature map/Grid → Set of slots)
+	
+
+- 参考论文：
+
+    @article{locatello2020object,
+    title={Object-Centric Learning with Slot Attention},
+    author={Locatello, Francesco and Weissenborn, Dirk and Unterthiner, Thomas and Mahendran, Aravindh and Heigold, Georg and Uszkoreit, Jakob and Dosovitskiy, Alexey and Kipf, Thomas},
+    journal={arXiv preprint arXiv:2006.15055},
+    year={2020}
+}
+
+- 参考实现：
+
+   https://github.com/google-research/google-research/tree/master/slot_attention
+
+- 适配昇腾 AI 处理器的实现：
+    
+https://gitee.com/ascend/ModelZoo-TensorFlow/tree/master/TensorFlow/contrib/cv/Slot-Attention_ID2028_for_TensorFlow
+
+<h2 id="概述.md">原始模型</h2>
+
+obs地址：obs://lwr-slot-npu/slottl/newslotmodel.pb
+
+
+步骤一:
+通过代码keras_frozen_graph将ckpt-499000转成pb
+ckpt的obs地址：obs://lwr-slot-npu/slottl/
+该目录中以checkpoint开头的四个文件
+
+
+<h2 id="概述.md">pb模型</h2>
+
+```
+newslotmodel.pb
+```
+pb文件的obs地址：obs://lwr-slot-npu/slottl/newslotmodel.pb
+
+
+<h2 id="概述.md">om模型</h2>
+
+转newslotmodel.pb到slotmodel.om
+
+使用ATC模型转换工具进行模型转换时可以参考如下指令:
+
+```
+atc --model=./newslotmodel.pb --input_shape="input:64, 128, 128, 3" --framework=3 --output=slotmodel --soc_version=Ascend910A --precision_mode=force_fp32 --op_select_implmode=high_precision
+```
+
+成功转化成slotmodel.om
+
+slotmodel.om的obs地址:obs://lwr-slot-npu/slottl/slotmodel.om
+
+
+
+<h2 id="概述.md">使用msame工具推理</h2>
+
+参考 https://gitee.com/ascend/tools/tree/master/msame, 获取msame推理工具及使用方法。
+
+使用msame推理工具，参考如下命令，发起推理测试：
+
+```
+./msame --model "slotmodel.om"  --output "./" --outfmt TXT --loop 1
+```
\ No newline at end of file
diff --git a/ACL_TensorFlow/contrib/cv/Slot-Attention_ID2028_for_ACL/keras_frozen_graph.py b/ACL_TensorFlow/contrib/cv/Slot-Attention_ID2028_for_ACL/keras_frozen_graph.py
new file mode 100644
index 0000000000000000000000000000000000000000..9a38dd9e520386d5d85318406a61826f550fc37f
--- /dev/null
+++ b/ACL_TensorFlow/contrib/cv/Slot-Attention_ID2028_for_ACL/keras_frozen_graph.py
@@ -0,0 +1,26 @@
+import tensorflow as tf
+from tensorflow.python.framework import graph_util
+import model as model_utils
+
+inputs = tf.placeholder(tf.float32, shape=[64, 128, 128, 3],name='input')  # input shape
+    # create inference graph
+resolution = (128, 128)
+batch_size=64
+num_slots=7
+num_iterations=3
+
+model = model_utils.build_model(resolution, batch_size, num_slots,
+                                num_iterations, model_type="object_discovery")
+logit1,logit2,logit3,logit4 = model(inputs, training=False)
+print("-----------------------------------------测试完成-------------------------------------------")
+saver = tf.train.Saver(max_to_keep=5)
+# graph_def = tf.get_default_graph().as_graph_def()
+
+with tf.Session() as sess:
+    saver.restore(sess, '/home/disk/checkp/checkpoint.ckpt-499000')
+    print("---------------------开始转换-------------------------------")
+    output_graph_def = graph_util.convert_variables_to_constants(sess,sess.graph_def,['model/slot_attention_auto_encoder/Sum'])
+    print("---------------------转换完成--------------------------")
+    # print(sess.run(tf.get_default_graph().get_tensor_by_name('model/slot_attention_auto_encoder/Sum:0')))  # 3.0
+    with tf.gfile.GFile('./newslotmodel.pb', 'wb') as f:
+        f.write(output_graph_def.SerializeToString())  # 得到文件：model.pb