diff --git a/models/cv/segmentation/mask_rcnn/ixrt/README.md b/models/cv/segmentation/mask_rcnn/ixrt/README.md
index 8b2bd2b7689b50dfdad9f526d80b880eb19eed70..ab86e0886d58ed6a8bd77e50eca92c68f3acafae 100644
--- a/models/cv/segmentation/mask_rcnn/ixrt/README.md
+++ b/models/cv/segmentation/mask_rcnn/ixrt/README.md
@@ -4,45 +4,60 @@
 
 Mask R-CNN (Mask Region-Based Convolutional Neural Network) is an extension of the Faster R-CNN model, which is itself an improvement over R-CNN and Fast R-CNN. Developed by Kaiming He et al., Mask R-CNN is designed for object instance segmentation tasks, meaning it not only detects objects within an image but also generates high-quality segmentation masks for each instance.
 
-## Setup
+## Prepare
+
+```bash
+# go to current model home path
+cd ${PROJ_ROOT}/models/cv/segmentation/mask_rcnn/ixrt
+```
+
+Prepare weights and datasets referring to below steps:
+
+"maskrcnn.wts" [export method](https://github.com/wang-xinyu/tensorrtx/tree/master/rcnn#how-to-run)
 
-Prepare on MR GPU
+- use the [script](https://github.com/wang-xinyu/tensorrtx/blob/master/rcnn/gen_wts.py)
+- use the [config file](https://github.com/facebookresearch/detectron2/blob/main/configs/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x.yaml)
+- use [weights](https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x/137259246/model_final_9243eb.pkl)
 
 ```bash
-cd PATH/TO/scripts
-bash init.sh
+# put maskrcnn.wts in "python/maskrcnn.wts"
+wget -P python/ http://files.deepspark.org.cn:880/deepspark/wts/maskrcnn.wts
 ```
 
-Prepare on NV GPU
+Visit [COCO site](https://cocodataset.org/) and get COCO2017 datasets
+
+- images directory: coco/images/val2017/*.jpg
+- annotations directory: coco/annotations/instances_val2017.json
+
+## Setup
 
 ```bash
-cd PATH/TO/scripts
-bash init_nv.sh
+cd scripts/
 ```
 
-### Model Conversion
+### Prepare on MR GPU
 
-Prepare weights and datasets referring to below steps:
+```bash
+bash init.sh
+```
+
+### Prepare on NV GPU
 
-- "maskrcnn.wts" [export method](https://github.com/wang-xinyu/tensorrtx/tree/master/rcnn#how-to-run), use the [script](https://github.com/wang-xinyu/tensorrtx/blob/master/rcnn/gen_wts.py), use the [config file](https://github.com/facebookresearch/detectron2/blob/main/configs/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x.yaml), use [weights](https://dl.fbaipublicfiles.com/detectron2/COCO-InstanceSegmentation/mask_rcnn_R_50_C4_1x/137259246/model_final_9243eb.pkl)
-  - put maskrcnn.wts in "python/maskrcnn.wts"
-- Visit [COCO site](https://cocodataset.org/) and get COCO2017 datasets
-  - images directory: coco/images/val2017/*.jpg
-  - annotations directory: coco/annotations/instances_val2017.json
+```bash
+bash init_nv.sh
+```
 
 ## Inference
 
 ### FP16 Performance
 
 ```bash
-cd PATH/TO/scripts
 bash infer_maskrcnn_fp16_performance.sh
 ```
 
 ### FP16 Accuracy
 
 ```bash
-cd PATH/TO/scripts
 bash infer_maskrcnn_fp16_acc.sh
 ```
 
diff --git a/models/cv/segmentation/mask_rcnn/ixrt/python/maskrcnn.py b/models/cv/segmentation/mask_rcnn/ixrt/python/maskrcnn.py
index 6362d01167e0f3a8490b9d16875a12d39d83c7b2..75484195b0424a3ef852e620fb1dea2390048700 100644
--- a/models/cv/segmentation/mask_rcnn/ixrt/python/maskrcnn.py
+++ b/models/cv/segmentation/mask_rcnn/ixrt/python/maskrcnn.py
@@ -206,12 +206,12 @@ def get_maskrcnn_perf(config):
 
     # Process I/O and execute the network
     cuda.memcpy_htod(inputs[0]["allocation"], data_batch)
-    # warm up 
+    # warm up
     print("Warmup start ...")
     for i in range(5):
         context.execute_v2(allocations)
     print("Warmup done !\nStart forward ...")
-    # run 
+    # run
     forward_time = 0
     for i in range(20):
         start_time = time.time()
@@ -268,12 +268,12 @@ def get_maskrcnn_acc(config):
     # Prepare the output data
     output = np.zeros(outputs[0]["shape"], outputs[0]["dtype"])
 
-    # warm up 
+    # warm up
     print("Warmup start ...")
     for i in range(3):
         context.execute_v2(allocations)
     print("Warmup done !\nStart forward ...")
-    
+
     # run
     start_time = time.time()
     for batch_data, batch_img_shape, batch_img_id, batched_paddings, paths in tqdm(dataloader):
@@ -287,7 +287,7 @@ def get_maskrcnn_acc(config):
         # cpu -> gpu
         batch_data = np.ascontiguousarray(batch_data)
         cuda.memcpy_htod(inputs[0]["allocation"], batch_data)
-        
+
         context.execute_v2(allocations)
 
         # gpu -> cpu
@@ -318,17 +318,17 @@ def get_maskrcnn_acc(config):
 
     print(F"E2E time : {end2end_time:.3f} seconds")
     print("Forward done !")
-    
+
     tmp_result_name = "pred_results.json"
     if os.path.exists(tmp_result_name):
         os.remove(tmp_result_name)
     with open(tmp_result_name, "w") as f:
         json.dump(json_result, f)
 
-    
+
     anno = COCO(anno_json)  # init annotations api
     pred = anno.loadRes(tmp_result_name)  # init predictions api
-    
+
     eval = COCOeval(anno, pred, "bbox")
     eval.evaluate()
     eval.accumulate()
@@ -344,11 +344,11 @@ def get_maskrcnn_acc(config):
     _, map50 = eval.stats[:2]
     print("bbox mAP@0.5 : ", map50)
     print(f"bbox Accuracy Check : Test {map50} >= target {config.map_target}")
-    
+
     _, segm_map50 = segm_eval.stats[:2]
     print("segm mAP@0.5 : ", segm_map50)
     print(f"segm Accuracy Check : Test {segm_map50} >= target {config.segm_map_target}")
-    
+
     if map50 >= config.map_target and segm_map50 >= config.segm_map_target:
         print("pass!")
     else:
diff --git a/models/cv/segmentation/mask_rcnn/ixrt/scripts/init.sh b/models/cv/segmentation/mask_rcnn/ixrt/scripts/init.sh
index 7b699195a5bdf40b74f2d610a51bb5b17e5415ef..bd1fe1177a797dc149968f82017ee25877667c00 100644
--- a/models/cv/segmentation/mask_rcnn/ixrt/scripts/init.sh
+++ b/models/cv/segmentation/mask_rcnn/ixrt/scripts/init.sh
@@ -14,13 +14,14 @@
 #    under the License.
 
 # build plugin
-cd ../plugins
+pushd ../plugins
 rm -rf build
 mkdir build
 cd build
-cmake .. -DIXRT_HOME=/usr/local/corex
-# cmake .. -DIXRT_HOME=/opt/sw_home/local
+cmake .. -D CMAKE_C_COMPILER="/usr/local/corex/bin/clang" -D CMAKE_CXX_COMPILER="/usr/local/corex/bin/clang++" -DIXRT_HOME=/usr/local/corex
+
 make -j8
+popd
 
 
 ## install packages
@@ -28,10 +29,10 @@ bash prepare_system_env.sh
 
 # pip whl
 #pip3 install opencv-python==4.6.0.66
-pip3 install pycocotools==2.0.2
+pip3 install pycocotools==2.0.7
 pip3 install tqdm
 
 # build engine
-cd ../../python
+cd ../python
 rm -rf ./maskrcnn.engine
 python3 maskrcnn.py build_engine --wts_file ./maskrcnn.wts --engine_file ./maskrcnn.engine