diff --git a/AscendIE/TorchAIE/built-in/cv/detection/YoloX/README.md b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..28badefec47da2b29bbea862467f6db7616bed15
--- /dev/null
+++ b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/README.md
@@ -0,0 +1,211 @@
+# YOLOX模型-AIE推理引擎部署指导
+
+- [概述](#ZH-CN_TOPIC_0000001172161501)
+
+ - [输入输出数据](#section540883920406)
+
+- [推理环境准备](#ZH-CN_TOPIC_0000001126281702)
+
+- [模型部署](#ZH-CN_TOPIC_0000001126281700)
+
+ - [安装依赖](#section4622531142816)
+ - [准备数据集](#section183221994411)
+ - [模型推理](#section741711594517)
+
+- [模型推理性能&精度](#ZH-CN_TOPIC_0000001172201573)
+
+******
+
+# 概述
+
+YOLOX是基于往年对YOLO系列众多改进而产生的目标检测模型，其采用无锚方式，并应用了解耦头和领先的标签分配策略 SimOTA.其在众多数据集中均获得了最佳结果。
+
+- 参考实现：
+
+ ```
+ url=https://github.com/Megvii-BaseDetection/YOLOX
+ commit_id=6880e3999eb5cf83037e1818ee63d589384587bd
+ code_path=ACL_PyTorch/contrib/cv/detection/YOLOX
+ model_name=YOLOX
+ ```
+
+## 输入输出数据
+
+- 输入数据
+
+ | 输入数据 | 数据类型 | 大小 | 数据排布格式 |
+ | -------- | -------- | ------------------------- | ------------ |
+ | input | RGB_FP32 | batchsize x 3 x 640 x 640 | NCHW |
+
+- 输出数据
+
+ | 输出数据 | 数据类型 | 大小 |
+ | -------- | -------- | ------- |
+ | output | FLOAT32 | 1 x 8400 x 85 |
+
+# 推理环境准备
+
+- 该模型需要以下插件与驱动
+
+ **表 1** 版本配套表
+
+ | 配套 | 版本 | 环境准备指导 |
+ | --------------------------------------------------------------- | ------- | ----------------------------------------------------------------------------------------------------- |
+ | 固件与驱动 | 23.0.rc1 | [Pytorch框架推理环境准备](https://www.hiascend.com/document/detail/zh/ModelZoo/pytorchframework/pies) |
+ | CANN | 7.0.RC1.alpha003 | - |
+ | Python | 3.9.11 | - |
+ | PyTorch | 2.0.1 | - |
+ | torch_aie | 6.3rc2 |
+ | 说明：Atlas 300I Duo 推理卡请以CANN版本选择实际固件与驱动版本。 | \ | \ |
+
+# 模型部署
+
+## 安装依赖
+
+1. 获取源码
+
+ ```shell
+ git clone https://github.com/Megvii-BaseDetection/YOLOX
+ cd YOLOX
+ git reset 6880e3999eb5cf83037e1818ee63d589384587bd --hard
+ patch -p1 < ../yolox_coco_evaluator.patch
+ pip install -v -e . # or python3 setup.py develop
+ cd ..
+ ```
+
+2. 安装需要的Python Library
+
+ ```shell
+ apt-get install libprotobuf-dev protobuf-compiler
+ apt-get install libgl1-mesa-glx
+ pip install -r requirements.txt
+ ```
+
+3. 设置环境变量
+
+ 执行环境中推理引擎安装路径下的环境变量设置脚本
+
+ ```shell
+ source {aie}/set_env.sh
+ ```
+
+## 准备数据集
+
+1. 获取原始数据集。（解压命令参考tar –xvf \*.tar与 unzip \*.zip）
+
+ 请参考开源代码仓方式获得[COCO2017数据集](https://cocodataset.org/)，并根据需要置于服务器上（如 `datasets_path=/data/dataset/coco`），val2017目录存放coco数据集的验证集图片，annotations目录存放coco数据集的instances_val2017.json，文件目录结构如下：
+
+ ```
+ data
+ ├── dataset
+ │ ├── coco
+ │ │ ├── annotations
+ │ │ ├── val2017
+ ```
+
+## 模型推理
+
+### 1. 模型转换
+
+ 使用PyTorch将模型权重文件.pth转换为torchscript文件
+
+ 1. 获取权重文件
+
+ 我们利用官方的PTH文件进行验证，官方PTH文件可从原始开源库中获取，我们需要[yolox_x.pth](https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_x.pth)文件,请将其放在与README.md文件同一目录内。
+
+ 2. 导出torchscript文件
+
+ ```shell
+ cd YOLOX
+ python tools/export_torchscript.py \
+ --output-name ../yolox.torchscript.pt \
+ -n yolox-x \
+ -c ../yolox_x.pth
+ cd ..
+ ```
+
+ 获得yolox.torchscript.pt文件。
+
+ + 参数说明
+ + `--output-name`：输出文件名称
+ + `-n`：模型名称
+ + `-c`：权重文件路径
+
+### 2. 开始推理验证
+
+ 1. 执行命令查看芯片名称（$\{chip\_name\}）。
+
+ ```shell
+ npu-smi info
+ #该设备芯片名为Ascend310P3 （在下一步中赋值给soc_version环境变量）
+ 回显如下：
+ +-------------------+-----------------+------------------------------------------------------+
+ | NPU Name | Health | Power(W) Temp(C) Hugepages-Usage(page) |
+ | Chip Device | Bus-Id | AICore(%) Memory-Usage(MB) |
+ +===================+=================+======================================================+
+ | 0 310P3 | OK | 15.8 42 0 / 0 |
+ | 0 0 | 0000:82:00.0 | 0 1074 / 21534 |
+ +===================+=================+======================================================+
+ | 1 310P3 | OK | 15.4 43 0 / 0 |
+ | 0 1 | 0000:89:00.0 | 0 1070 / 21534 |
+ +===================+=================+======================================================+
+ ```
+
+ 2. 对原生ts文件执行torch_aie编译，导出NPU支持的ts文件
+
+ ```shell
+ soc_version="Ascend310P3" # User-defined
+ python yolox_export_torch_aie_ts.py \
+ --torch-script-path ./yolox.torchscript.pt \
+ --batch-size 1 \
+ --save-path ./yoloxb1_torch_aie.pt \
+ --soc-version ${soc_version}
+ ```
+ + 参数说明
+ + `--torch-script-path`：原生ts文件路径
+ + `--batch-size`：用户自定义的batch size
+ + `--save-path`：AIE编译后的ts文件保存路径
+ + `--soc-version`：NPU型号
+
+ 3. 执行推理并验证精度与性能
+
+ COCO2017数据集需要约5分钟完成所有推理任务，请耐心等待。
+
+ ```shell
+ python yolox_eval.py \
+ --dataroot /data/dataset/coco \
+ --batch 1 \
+ --ts ./yoloxb1_torch_aie.pt
+ ```
+
+ - 参数说明：
+ - --dataroot：COCO数据集的路径，同上
+ - --batch：用户自定义的batch size
+ - --ts：AIE编译后的ts文件路径
+
+ 运行成功后将打印该模型在NPU推理结果的精度信息与性能信息。
+
+
+# 模型推理性能&精度
+
+基于推理引擎完成推理计算，精度与性能可参考下列数据：
+
+| Soc version | Batch Size | Dataset | Accuracy ||
+| ---------- | ---------- | ---------- | ---------- | ---------- |
+| 310P3精度 | 1 | coco2017 | Average Precision(IoU=0.50:0.95): 0.498 | Average Precision(IoU=0.50): 0.678 |
+| 310P3精度 | 4 | coco2017 | Average Precision(IoU=0.50:0.95): 0.498 | Average Precision(IoU=0.50): 0.678 |
+| 310P3精度 | 8 | coco2017 | Average Precision(IoU=0.50:0.95): 0.498 | Average Precision(IoU=0.50): 0.678 |
+| 310P3精度 | 20 | coco2017 | Average Precision(IoU=0.50:0.95): 0.498 | Average Precision(IoU=0.50): 0.678 |
+
+| Soc version | Batch Size | Dataset | Performance |
+| -------- | ---------- | ---------- | ---------- |
+| 310P3 | 1 | coco2017 | 35.65 ms/pic |
+| 310P3 | 4 | coco2017 | 33.09 ms/pic |
+| 310P3 | 8 | coco2017 | 33.63 ms/pic |
+| 310P3 | 20 | coco2017 | 32.67 ms/pic |
+
+# FAQ
+1. 若遇到类似报错：ImportError: /lib/aarch64-linux-gnu/libGLdispatch.so.0: cannot allocate memory in static TLS block
+
+ 解决方法：
+ export LD_PRELOAD=$LD_PRELOAD:{报错信息中的路径}
\ No newline at end of file
diff --git a/AscendIE/TorchAIE/built-in/cv/detection/YoloX/requirements.txt b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/requirements.txt
new file mode 100644
index 0000000000000000000000000000000000000000..92675b2315a04ff5dea7c40fbea40aac2396b4f3
--- /dev/null
+++ b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/requirements.txt
@@ -0,0 +1,12 @@
+numpy
+torch==2.0.1
+opencv_python
+loguru
+scikit-image
+tqdm
+torchvision
+Pillow
+thop
+ninja
+tabulate
+tensorboard
diff --git a/AscendIE/TorchAIE/built-in/cv/detection/YoloX/test.sh b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/test.sh
new file mode 100644
index 0000000000000000000000000000000000000000..7f767ca7cd13c27aebea54ff634cd7a787d63980
--- /dev/null
+++ b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/test.sh
@@ -0,0 +1,62 @@
+# Copyright(C) 2023. Huawei Technologies Co.,Ltd. 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.
+
+# Description: Test script for running YoloX model with Ascend Inference Engine
+# Author: chenchuwei c00818886
+# Create: 2023/10/20
+
+soc_version="Ascend310P3"
+data_root="/data/dataset/coco"
+batch=1
+
+if [ ! -f "yolox_x.pth" ]; then
+ echo "[ERROR] yolox_x.pth not found in current dir, please make sure it exists."
+ exit 1
+fi
+
+if [ ! -d "YOLOX" ]; then
+ echo "[INFO] Preparing YoloX's dependencies"
+ git clone https://github.com/Megvii-BaseDetection/YOLOX
+ cd YOLOX
+ git reset 6880e3999eb5cf83037e1818ee63d589384587bd --hard
+ patch -p1 < ../yolox_coco_evaluator.patch
+ pip install -v -e .
+ cd ..
+fi
+
+echo "[INFO] Installing Python dependencies"
+apt-get install libprotobuf-dev protobuf-compiler -y
+apt-get install libgl1-mesa-glx -y
+pip install -r requirements.txt
+
+if [ ! -f "yolox.torchscript.pt" ]; then
+ echo "[INFO] Exporting torchscript module"
+ cd YOLOX
+ python tools/export_torchscript.py --output-name ../yolox.torchscript.pt -n yolox-x -c ../yolox_x.pth
+ cd ..
+fi
+
+if [ ! -f "yoloxb${batch}_torch_aie.pt" ]; then
+ echo "[INFO] AIE Compiling"
+ python yolox_export_torch_aie_ts.py \
+ --torch-script-path ./yolox.torchscript.pt \
+ --batch-size ${batch} \
+ --soc-version ${soc_version}
+fi
+
+echo "[INFO] Start AIE evaluation"
+python yolox_eval.py \
+ --dataroot ${data_root} \
+ --batch ${batch} \
+ --ts ./yoloxb${batch}_torch_aie.pt
\ No newline at end of file
diff --git a/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_coco_evaluator.patch b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_coco_evaluator.patch
new file mode 100755
index 0000000000000000000000000000000000000000..a81cb0971ec7437cbc1d680d96a167fe92148820
--- /dev/null
+++ b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_coco_evaluator.patch
@@ -0,0 +1,67 @@
+diff --git a/yolox/evaluators/coco_evaluator.py b/yolox/evaluators/coco_evaluator.py
+index 96eb56a..64827fc 100644
+--- a/yolox/evaluators/coco_evaluator.py
++++ b/yolox/evaluators/coco_evaluator.py
+@@ -164,33 +164,34 @@ class COCOEvaluator:
+ data_list.append(pred_data)
+ return data_list
+
+- def evaluate_prediction(self, data_dict, statistics):
++ def evaluate_prediction(self, data_dict, statistics=None):
+ if not is_main_process():
+ return 0, 0, None
+
+ logger.info("Evaluate in main process...")
+
+ annType = ["segm", "bbox", "keypoints"]
++
++ if statistics is not None:
++ inference_time = statistics[0].item()
++ nms_time = statistics[1].item()
++ n_samples = statistics[2].item()
++
++ a_infer_time = 1000 * inference_time / (n_samples * self.dataloader.batch_size)
++ a_nms_time = 1000 * nms_time / (n_samples * self.dataloader.batch_size)
++
++ time_info = ", ".join(
++ [
++ "Average {} time: {:.2f} ms".format(k, v)
++ for k, v in zip(
++ ["forward", "NMS", "inference"],
++ [a_infer_time, a_nms_time, (a_infer_time + a_nms_time)],
++ )
++ ]
++ )
+
+- inference_time = statistics[0].item()
+- nms_time = statistics[1].item()
+- n_samples = statistics[2].item()
+-
+- a_infer_time = 1000 * inference_time / (n_samples * self.dataloader.batch_size)
+- a_nms_time = 1000 * nms_time / (n_samples * self.dataloader.batch_size)
+-
+- time_info = ", ".join(
+- [
+- "Average {} time: {:.2f} ms".format(k, v)
+- for k, v in zip(
+- ["forward", "NMS", "inference"],
+- [a_infer_time, a_nms_time, (a_infer_time + a_nms_time)],
+- )
+- ]
+- )
+-
+- info = time_info + "\n"
+-
++ info = time_info + "\n"
++ info = "\n"
+ # Evaluate the Dt (detection) json comparing with the ground truth
+ if len(data_dict) > 0:
+ cocoGt = self.dataloader.dataset.coco
+@@ -216,6 +217,7 @@ class COCOEvaluator:
+ with contextlib.redirect_stdout(redirect_string):
+ cocoEval.summarize()
+ info += redirect_string.getvalue()
+- return cocoEval.stats[0], cocoEval.stats[1], info
++ return info
+ else:
+ return 0, 0, info
diff --git a/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_eval.py b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_eval.py
new file mode 100644
index 0000000000000000000000000000000000000000..3dd182eca6960be01e02814e9260cb55fbdf6241
--- /dev/null
+++ b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_eval.py
@@ -0,0 +1,96 @@
+# Copyright(C) 2023. Huawei Technologies Co.,Ltd. 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 os
+import sys
+import argparse
+import time
+import torch
+import torch_aie
+import numpy as np
+
+from tqdm import tqdm
+from ais_bench.infer.interface import InferSession
+from yolox.data import COCODataset, ValTransform
+from yolox.evaluators import COCOEvaluator
+from yolox.utils.boxes import postprocess
+from yolox.utils.demo_utils import demo_postprocess
+
+
+def main():
+ print("[INFO] YoloX AIE evaluation process start")
+
+ parser = argparse.ArgumentParser(description="YOLOX Preprocess")
+ parser.add_argument('--dataroot', dest='dataroot',
+ help='data root dirname', default='/data/datasets/coco',
+ type=str)
+ parser.add_argument('--batch',
+ help='validation batch size', default=1,
+ type=int)
+ parser.add_argument('--ts',
+ help='root of ts module', default="./yoloxb1_torch_aie.pt",
+ type=str)
+ opt = parser.parse_args()
+
+ valdataset = COCODataset(
+ data_dir=opt.dataroot,
+ json_file='instances_val2017.json',
+ name="val2017",
+ img_size=(640, 640),
+ preproc=ValTransform(legacy=False),
+ )
+ sampler = torch.utils.data.SequentialSampler(valdataset)
+
+ dataloader_kwargs = {
+ "num_workers": 8, "pin_memory": True, "sampler": sampler, "batch_size": opt.batch
+ }
+
+ val_loader = torch.utils.data.DataLoader(valdataset, **dataloader_kwargs)
+
+ torch_aie.set_device(0)
+
+ print("[INFO] Loading TS module")
+ aie_module = torch.jit.load(opt.ts)
+ aie_module.eval()
+
+ aie_cost = 0
+ infer_times = 0
+ data_list = []
+ coco_evaluator = COCOEvaluator(val_loader, img_size=(640, 640), confthre=0.001, nmsthre=0.65, num_classes=80)
+
+ print(f"[INFO] Start AIE inference, please be patient (batch={opt.batch})")
+ for _, datas in enumerate(tqdm(val_loader)):
+ data = datas[0]
+
+ # Inference with AIE-compiled TS module
+ start = time.time()
+ result = aie_module(data)
+ cost = time.time() - start
+
+ outputs = demo_postprocess(result, [640, 640])
+ outputs = postprocess(outputs, num_classes=80, conf_thre=0.001, nms_thre=0.65)
+
+ data_list.extend(coco_evaluator.convert_to_coco_format(outputs, datas[2], datas[3]))
+ aie_cost += cost
+ infer_times += 1
+
+ # Use COCO_Evaluator to evaluate the accuracy
+ coco_result = coco_evaluator.evaluate_prediction(data_list)
+ print(coco_result)
+ print(f'\n[INFO] PT-AIE inference avg cost: {aie_cost / infer_times * 1000 / opt.batch} ms/pic')
+ print(f'[INFO] Total sample count = {infer_times * opt.batch} pics')
+ print('[INFO] YoloX AIE evaluation process finished')
+
+if __name__ == "__main__":
+ main()
diff --git a/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_export_torch_aie_ts.py b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_export_torch_aie_ts.py
new file mode 100644
index 0000000000000000000000000000000000000000..8c1ca305ce01f5bce04dfca6c2d548ae2480baae
--- /dev/null
+++ b/AscendIE/TorchAIE/built-in/cv/detection/YoloX/yolox_export_torch_aie_ts.py
@@ -0,0 +1,57 @@
+# Copyright(C) 2023. Huawei Technologies Co.,Ltd. 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 sys
+import os
+import argparse
+import torch
+import torch_aie
+from torch_aie import _enums
+
+
+def export_torch_aie(model_path, batch_size, soc_version, save_path="./"):
+ trace_model = torch.jit.load(model_path)
+ trace_model.eval()
+ input_info = [torch_aie.Input((batch_size, 3, 640, 640))]
+ torch_aie.set_device(0)
+ torchaie_model = torch_aie.compile(
+ trace_model,
+ inputs=input_info,
+ torch_executed_ops=[],
+ precision_policy=_enums.PrecisionPolicy.FP32,
+ soc_version=soc_version,
+ )
+ suffix = os.path.splitext(model_path)[-1]
+ saved_name = os.path.basename(model_path).split('.')[0] + f"b{batch_size}_torch_aie" + suffix
+ torchaie_model.save(os.path.join(save_path, saved_name))
+ print("[INFO] torch_aie compile for YoloX finished, model saved in: ", os.path.join(save_path, saved_name))
+
+
+def parse_opt():
+ parser = argparse.ArgumentParser()
+ parser.add_argument('--torch-script-path', type=str, default='./yolox.torchscript.pt', help='trace model path')
+ parser.add_argument('--batch-size', type=int, default=1, help='batch size')
+ parser.add_argument('--save-path', type=str, default='./', help='compiled model path')
+ parser.add_argument('--soc-version', type=str, default='Ascend310P3', help='soc version')
+ opt_args = parser.parse_args()
+ return opt_args
+
+
+def main(opt_args):
+ print("[INFO] torch_aie compile for YoloX start")
+ export_torch_aie(opt_args.torch_script_path, opt_args.batch_size, opt_args.soc_version, opt_args.save_path)
+
+if __name__ == '__main__':
+ opt = parse_opt()
+ main(opt)
\ No newline at end of file