diff --git a/configs/DensePose_ResNet101_FPN_32x8d_s1x-e2e.yaml b/configs/DensePose_ResNet101_FPN_32x8d_s1x-e2e.yaml
index 53bf5f6..0968913 100644
--- a/configs/DensePose_ResNet101_FPN_32x8d_s1x-e2e.yaml
+++ b/configs/DensePose_ResNet101_FPN_32x8d_s1x-e2e.yaml
@@ -76,12 +76,12 @@ TRAIN:
   USE_FLIPPED: True
 TEST:
   DATASETS: ('dense_coco_2014_minival',)
-  PROPOSAL_LIMIT: 1000
+  PROPOSAL_LIMIT: 2000
   SCALE: 800
   MAX_SIZE: 1333
   NMS: 0.5
   FORCE_JSON_DATASET_EVAL: True
   DETECTIONS_PER_IM: 20
-  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
-  RPN_POST_NMS_TOP_N: 1000
+  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
+  RPN_POST_NMS_TOP_N: 2000
 OUTPUT_DIR: ''
diff --git a/deploy.py b/deploy.py
new file mode 100644
index 0000000..b0b8880
--- /dev/null
+++ b/deploy.py
@@ -0,0 +1,139 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+from collections import defaultdict
+import argparse
+import cv2  # NOQA (Must import before importing caffe2 due to bug in cv2)
+import glob
+import logging
+import os
+import sys
+import time
+
+from caffe2.python import workspace
+import numpy as np
+from detectron.core.config import assert_and_infer_cfg
+from detectron.core.config import cfg
+from detectron.core.config import merge_cfg_from_file
+from detectron.utils.io import cache_url
+from detectron.utils.logging import setup_logging
+from detectron.utils.timer import Timer
+import detectron.core.test_engine as infer_engine
+import detectron.datasets.dummy_datasets as dummy_datasets
+import detectron.utils.c2 as c2_utils
+import detectron.utils.vis as vis_utils
+
+c2_utils.import_detectron_ops()
+
+# OpenCL may be enabled by default in OpenCV3; disable it because it's not
+# thread safe and causes unwanted GPU memory allocations.
+cv2.ocl.setUseOpenCL(False)
+
+workspace.GlobalInit(['caffe2', '--caffe2_log_level=0'])
+setup_logging(__name__)
+
+
+##############################
+###        Settings        ###
+##############################
+parser = argparse.ArgumentParser()
+parser.add_argument('--gpu_id', type=str, default='0', help='gpu id: e.g. 0 1 2 3')
+parser.add_argument('--host', type=str, default='0.0.0.0')
+parser.add_argument('--port', type=int, default=5000)
+args = parser.parse_args()
+os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
+
+
+##############################
+###         flask          ###
+##############################
+from flask import Flask
+from flask_jsonrpc import JSONRPC
+
+# Flask application
+app = Flask(__name__)
+app.config['JSON_AS_ASCII'] = False
+
+# Flask-JSONRPC
+jsonrpc = JSONRPC(app, '/api', enable_web_browsable_api=False)
+
+
+##############################
+###          model         ###
+##############################
+# logger = logging.getLogger(__name__)
+merge_cfg_from_file('./configs/DensePose_ResNet101_FPN_32x8d_s1x-e2e.yaml')
+cfg.NUM_GPUS = 1
+assert_and_infer_cfg(cache_urls=False)
+DOWNLOAD_CACHE = './detectron-download-cache/DensePose_ResNet101_FPN_32x8d_s1x-e2e.pkl'
+model = infer_engine.initialize_model_from_cfg(DOWNLOAD_CACHE)
+
+
+##############################
+###      detect_image      ###
+##############################
+def detect_image(im_path, thresh=0.9):
+    im = cv2.imread(im_path)
+    timers = defaultdict(Timer)
+    t = time.time()
+    with c2_utils.NamedCudaScope(0):
+        cls_boxes, cls_segms, cls_keyps, cls_bodys = infer_engine.im_detect_all(
+            model, im[:,:,::-1], None, timers=timers)
+        
+    if isinstance(cls_boxes, list):
+        boxes, segms, keypoints, classes = vis_utils.convert_from_cls_format(
+            cls_boxes, cls_segms, cls_keyps)
+
+    if boxes is None or boxes.shape[0] == 0 or max(boxes[:, 4]) < thresh:
+        return im, np.zeros(im.shape), np.zeros([im.shape[0],im.shape[1]])
+    # IUV
+    IUV_fields = cls_bodys[1]
+    # init 
+    All_Coords = np.zeros(im.shape)
+    All_inds = np.zeros([im.shape[0],im.shape[1]])
+    K = 26
+    ## inds
+    inds = np.argsort(-boxes[:,4])
+    ## 
+    for i, ind in enumerate(inds):
+        entry = boxes[ind,:]
+        if entry[4] > thresh:
+            entry=entry[0:4].astype(int)
+            ####
+            output = IUV_fields[ind]
+            ####
+            All_Coords_Old = All_Coords[ entry[1] : entry[1]+output.shape[1],entry[0]:entry[0]+output.shape[2],:]
+            All_Coords_Old[All_Coords_Old==0]=output.transpose([1,2,0])[All_Coords_Old==0]
+            All_Coords[ entry[1] : entry[1]+output.shape[1],entry[0]:entry[0]+output.shape[2],:]= All_Coords_Old
+            ###
+            CurrentMask = (output[0,:,:]>0).astype(np.float32)
+            All_inds_old = All_inds[ entry[1] : entry[1]+output.shape[1],entry[0]:entry[0]+output.shape[2]]
+            All_inds_old[All_inds_old==0] = CurrentMask[All_inds_old==0]*(i+1)
+            All_inds[ entry[1] : entry[1]+output.shape[1],entry[0]:entry[0]+output.shape[2]] = All_inds_old
+    # Result processing
+    All_Coords[:,:,1:3] = 255. * All_Coords[:,:,1:3]
+    All_Coords[All_Coords>255] = 255.
+    All_Coords = All_Coords.astype(np.uint8)
+    All_inds = All_inds.astype(np.uint8)
+    return im, All_Coords, All_inds
+
+##############################
+###          main          ###
+##############################
+# jsonrpc method
+@jsonrpc.method('main')
+def main(img_dir, output_dir, img_format='png', thresh=0.9):
+    # get img paths
+    img_paths = sorted(glob.glob(os.path.join(img_dir,'*'+img_format)))
+    print('image nums:', len(img_paths))
+    # detect
+    for img_path in tqdm(img_paths):
+        im, im_IUV, im_INDS = detect_image(img_path, thresh)
+        idx = img_path.split('/')[-1].split('_')[-1].split('.')[0]
+        output_path = os.path.join(output_dir, 'densepose'+'_'+idx+'.png')
+        cv2.imwrite(output_path, im_IUV)
+
+if __name__ == '__main__':
+    app.run(host=args.host, port=args.port, debug=True)
diff --git a/detectron/utils/vis.py b/detectron/utils/vis.py
index 31b2e79..e154282 100644
--- a/detectron/utils/vis.py
+++ b/detectron/utils/vis.py
@@ -384,11 +384,11 @@ def vis_one_image(
     All_inds = np.zeros([im.shape[0],im.shape[1]])
     K = 26
     ##
-    inds = np.argsort(boxes[:,4])
+    inds = np.argsort(-boxes[:,4])
     ##
     for i, ind in enumerate(inds):
         entry = boxes[ind,:]
-        if entry[4] > 0.65:
+        if entry[4] >= thresh:
             entry=entry[0:4].astype(int)
             ####
             output = IUV_fields[ind]
@@ -399,7 +399,7 @@ def vis_one_image(
             ###
             CurrentMask = (output[0,:,:]>0).astype(np.float32)
             All_inds_old = All_inds[ entry[1] : entry[1]+output.shape[1],entry[0]:entry[0]+output.shape[2]]
-            All_inds_old[All_inds_old==0] = CurrentMask[All_inds_old==0]*i
+            All_inds_old[All_inds_old==0] = CurrentMask[All_inds_old==0]*(i+1)
             All_inds[ entry[1] : entry[1]+output.shape[1],entry[0]:entry[0]+output.shape[2]] = All_inds_old
     #
     All_Coords[:,:,1:3] = 255. * All_Coords[:,:,1:3]