diff --git a/README.md b/README.md
new file mode 100644
index 0000000..4cef6b9
--- /dev/null
+++ b/README.md
@@ -0,0 +1,71 @@
+# ICL-Gait Dataset 1.0 
+Release Date: Sept 2021 
+
+## Primary Contact
+**Xiao Gu**, xiao.gu17@imperial.ac.uk Imperial College London
+
+**Yao Guo**, yao.guo@sjtu.edu.cn Shanghai Jiao Tong University
+
+## Citing
+Please cite the following paper for the use of this dataset
+
+Gu, Xiao, et al. "Occlusion-Invariant Rotation-Equivariant Semi-Supervised Depth Based Cross-View Gait Pose Estimation." arXiv preprint arXiv:2109.01397 (2021).
+
+## Supplementary Code
+* `vis_demo` provides script to visualize the data from different modalities
+* `syn` provides script to generate synthetic data based on SMPL
+
+## Dataset Details
+The dataset contains a real-world gait dataset collected from multiple viewpoints.
+Please see our [paper](https://arxiv.org/pdf/2109.01397) and the [website](https://xiaogu.site/ICL_gait) for more details. 
+
+## Data-Split Experiment Settings
+Please follow the settings in our [paper](https://arxiv.org/pdf/2109.01397) to benchmark your algorithms.
+
+### Cross-Subject (CS) Validation (2 loops)
+For cross-subject validation, the data were split to two groups {S01, S02, S04, S07}, {S03, S05, S06, S8}. 
+
+Each loop, use one group as training, and the other group as testing set. 
+
+### Cross-View (CV) Validation (5 loops)
+For cross-view validation, the data were split based on the five views. 
+
+Each loop, use the data from one view as training, the data from the other views as testing.  
+### Cross-Subject Cross-View (CS-CV) Validation (10 loops)
+For cross-subject & cross-view validation, the data was split to ten subgroups as a combination of CS and CV validation. 
+
+For example, one group is {S01-V01, S02-V01, S04-V01, S07-V01}. In each loop, use one group as the training set, and report the results on the remaining nine groups.
+
+**You can further split some proportion from the training set as a validation set, but any use of the testing data during training is not allowed.**
+
+
+## Folder Details
+### Dataset folder format
+S##_V##_C## refers to the data of the trial per subject, condition, and viewpoint. 
+S##: subject id
+V##: viewpoint 
+C##: walking condition
+
+Each folder contains 300 consecutive samples from one trial (the remaining samples leading to a much large data volume will be released in the future). 
+Missing trials (S1-C1-V1, S1-C2-V2, S1-C4-V1, S3-C3-V3, S5-C4-V4, S8-C2-V3, S8-C2-V4, S8-C4-V3, S8-C5-V3)
+
+* **depth**: contains the depth images recorded by RealSense D435
+  
+  ```
+  scale = 0.0010000000474974513;
+  fx = 928.108;  fy = 927.443;
+  cx = 647.394;  cy = 361.699
+  ```
+* **mask**:  contains the segmentation mask predicted from RGB images (access suspended) by CDCL
+  ```
+  ROI (lower-limb) RGB Value [255,127,127; 0,127,255; 0,0,255; 255,255,127; 127,255,127; 0,255,0]
+  ```
+* **point cloud**:  contains the point cloud converted from depth data, corresponding 3D keypoint, and root orientation
+  
+* **pose_2d**:  contains the 2D keypoints predicted by OpenPose
+
+* **kinematics**: contains the kinematics (randomly picked, not synchronized with the modalities above) which can be used synthetic data generation
+
+
+
+
diff --git a/syn/README.md b/syn/README.md
new file mode 100644
index 0000000..9c5c062
--- /dev/null
+++ b/syn/README.md
@@ -0,0 +1,19 @@
+# ICL-Gait to SMPL
+
+We provide the script [utils.py](smpl_webuser/utils.py) for transferring the gait parameter from our dataset to SMPL
+
+## Usage 
+* Please download the repository of [SMPL](smpl_webuser/utils.py) 
+
+* Add `models/pose.mat` to `models` of SMPL respository
+
+* Add `smpl_webuser/hello_world/demo_gait.py` to `smpl_webuser/hello_world/`
+
+* (optional) replace the `smpl_webuser/serialization.py` by a modified `smpl_webuser/serialization.py` compatible with python3
+
+* Running `demo_gait.py`, a synthetic model will be generated automatically
+
+
+
+
+
diff --git a/syn/models/pose.mat b/syn/models/pose.mat
new file mode 100644
index 0000000..4f3f6e8
Binary files /dev/null and b/syn/models/pose.mat differ
diff --git a/syn/smpl_webuser/hello_world/demo_gait.py b/syn/smpl_webuser/hello_world/demo_gait.py
new file mode 100644
index 0000000..e2e75e8
--- /dev/null
+++ b/syn/smpl_webuser/hello_world/demo_gait.py
@@ -0,0 +1,32 @@
+'''
+Modified from hellp_smpl.py with mocap2smpl function added
+'''
+
+from smpl_webuser.serialization import load_model
+import numpy as np
+from smpl_webuser.utils import mocap2smpl
+
+## Load SMPL model (here we load the female model)
+## Make sure path is correct
+m = load_model( '../../models/basicmodel_m_lbs_10_207_0_v1.1.0.pkl' )
+
+## Assign random pose and shape parameters
+# m.pose[:] = np.random.rand(m.pose.size) * .2
+m.betas[:] = np.random.rand(m.betas.size) * .03
+
+## assign poses from ICL
+import scipy.io as sio
+data = sio.loadmat('../../models/pose.mat')['pose'][0]
+mocap2smpl(data, m)
+
+## Write to an .obj file
+outmesh_path = './gait_smpl.obj'
+with open( outmesh_path, 'w') as fp:
+    for v in m.r:
+        fp.write( 'v %f %f %f\n' % ( v[0], v[1], v[2]) )
+
+    for f in m.f+1: # Faces are 1-based, not 0-based in obj files
+        fp.write( 'f %d %d %d\n' %  (f[0], f[1], f[2]) )
+
+## Print message
+print '..Output mesh saved to: ', outmesh_path
diff --git a/syn/smpl_webuser/serialization_new.py b/syn/smpl_webuser/serialization_new.py
new file mode 100644
index 0000000..e69de29
diff --git a/syn/smpl_webuser/utils.py b/syn/smpl_webuser/utils.py
new file mode 100644
index 0000000..290f46b
--- /dev/null
+++ b/syn/smpl_webuser/utils.py
@@ -0,0 +1,59 @@
+import transforms3d
+import numpy as np
+
+def mocap2smpl(data, m, global_rotation=[0,0,0.00001]):
+    ### global transform
+    p = data[89,90,91] # Pelvis
+    p_eulr = transforms3d.euler.axangle2euler(p/np.linalg.norm(p), np.linalg.norm(p)/180*np.pi, 'sxyz')
+    p_transform_woz = transforms3d.euler.euler2mat(p_eulr[0], p_eulr[1], 0)
+    p_transform = transforms3d.axangles.axangle2mat(p/np.linalg.norm(p), np.linalg.norm(p)/180*np.pi)
+    global_transform = transforms3d.axangles.axangle2mat(global_rotation, np.linalg.norm(global_rotation))
+    body_rotation = transforms3d.axangles.mat2axangle(np.dot(global_transform, p_transform_woz))
+    m.pose[0:3] = body_rotation[0][[1, 2, 0]]*body_rotation[1]
+
+    ### Left_UpperLeg transform
+    lfe = data[17:20] # LFemur
+    lfe_transform = transforms3d.axangles.axangle2mat(lfe/np.linalg.norm(lfe), np.linalg.norm(lfe)/180*np.pi)
+    lhip_transform = np.dot(np.linalg.inv(p_transform), lfe_transform) # relative rotation
+    lhip = transforms3d.axangles.mat2axangle(lhip_transform)
+    m.pose[3:6] = lhip[0][[1, 2, 0]]*lhip[1]
+
+    ### Right_UpperLeg transform
+    rfe = data[113:116] # RFemur
+    rfe_transform = transforms3d.axangles.axangle2mat(rfe/np.linalg.norm(rfe), np.linalg.norm(rfe)/180*np.pi)
+    rhip_transform = np.dot(np.linalg.inv(p_transform), rfe_transform)
+    rhip = transforms3d.axangles.mat2axangle(rhip_transform)
+    m.pose[6:9] = rhip[0][[1, 2, 0]]*rhip[1]
+
+    ### Left_LowerLeg transform
+    lti = data[71:74] # LTibia
+    lti_transform = transforms3d.axangles.axangle2mat(lti/np.linalg.norm(lti), np.linalg.norm(lti)/180*np.pi)
+    lk_transform = np.dot(np.linalg.inv(lfe_transform), lti_transform)
+    lk = transforms3d.axangles.mat2axangle(lk_transform)
+    m.pose[12:15] = lk[0][[1, 2, 0]]*lk[1]
+
+    ### Right_LowerLeg transform
+    rti = data[167:170] # RTibia
+    rti_transform = transforms3d.axangles.axangle2mat(rti/np.linalg.norm(rti), np.linalg.norm(rti)/180*np.pi)
+    rk_transform = np.dot(np.linalg.inv(rfe_transform), rti_transform)
+    rk = transforms3d.axangles.mat2axangle(rk_transform)
+    m.pose[15:18] = rk[0][[1, 2, 0]]*rk[1]
+
+    ### Left_Foot (90 offset)
+    lfo = data[26:29]
+    lfo_transform = transforms3d.axangles.axangle2mat(lfo/np.linalg.norm(lfo), np.linalg.norm(lfo)/180*np.pi)
+    lfo_transform = np.dot(lfo_transform, transforms3d.axangles.axangle2mat([0,1,0], np.pi/2))
+    la_transform = np.dot(np.linalg.inv(lti_transform), lfo_transform)
+    la = transforms3d.axangles.mat2axangle(la_transform)
+    m.pose[21:24] = la[0][[1, 2, 0]]*la[1]
+
+    ### Right_Foot (90 offset)
+    rfo = data[122:125]
+    rfo_transform = transforms3d.axangles.axangle2mat(rfo/np.linalg.norm(lfo), np.linalg.norm(rfo)/180*np.pi)
+    rfo_transform = np.dot(rfo_transform, transforms3d.axangles.axangle2mat([0,1,0], np.pi/2))
+    ra_transform = np.dot(np.linalg.inv(rti_transform), rfo_transform)
+    ra = transforms3d.axangles.mat2axangle(ra_transform)
+    m.pose[24:27] = ra[0][[1, 2, 0]]*ra[1]
+
+    ###
+
diff --git a/vis_demo/README.md b/vis_demo/README.md
new file mode 100644
index 0000000..53f3fb7
--- /dev/null
+++ b/vis_demo/README.md
@@ -0,0 +1,5 @@
+# Code for visualization data from different modalities
+
+Run `visualize.m` to visualize the data. 
+
+![alt text](demo_misc.png)
diff --git a/vis_demo/demo_misc.png b/vis_demo/demo_misc.png
new file mode 100644
index 0000000..2a6bf81
Binary files /dev/null and b/vis_demo/demo_misc.png differ
diff --git a/vis_demo/depth.png b/vis_demo/depth.png
new file mode 100644
index 0000000..13e0f35
Binary files /dev/null and b/vis_demo/depth.png differ
diff --git a/vis_demo/keypoint.json b/vis_demo/keypoint.json
new file mode 100644
index 0000000..56f0dc4
--- /dev/null
+++ b/vis_demo/keypoint.json
@@ -0,0 +1 @@
+{"version":1.3,"people":[{"person_id":[-1],"pose_keypoints_2d":[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,758.695,170.933,0.461265,786.089,174.877,0.409076,758.723,351.216,0.728424,774.32,515.742,0.714466,733.233,165.084,0.390594,652.908,333.579,0.693749,617.664,507.932,0.813623,0,0,0,0,0,0,0,0,0,0,0,0,551.009,511.862,0.657727,564.709,523.518,0.702532,627.433,525.489,0.676285,723.417,537.273,0.355216,729.271,527.554,0.331402,784.131,533.386,0.575139],"face_keypoints_2d":[],"hand_left_keypoints_2d":[],"hand_right_keypoints_2d":[],"pose_keypoints_3d":[],"face_keypoints_3d":[],"hand_left_keypoints_3d":[],"hand_right_keypoints_3d":[]}]}
\ No newline at end of file
diff --git a/vis_demo/mask.png b/vis_demo/mask.png
new file mode 100644
index 0000000..bd06238
Binary files /dev/null and b/vis_demo/mask.png differ
diff --git a/vis_demo/pointcloud.mat b/vis_demo/pointcloud.mat
new file mode 100644
index 0000000..8d34385
Binary files /dev/null and b/vis_demo/pointcloud.mat differ
diff --git a/vis_demo/visualize.m b/vis_demo/visualize.m
new file mode 100644
index 0000000..1d0c85e
--- /dev/null
+++ b/vis_demo/visualize.m
@@ -0,0 +1,62 @@
+% this is the script used for visualizing samples from ICL-Gait Dataset
+% https://xiaogu.site/ICL_gait/
+% Contact: xiao.gu17@imperial.ac.uk 
+
+clear all;
+close all;
+clc;
+
+figure();
+%% visualize depth
+depth = imread('depth.png');
+subplot(2,2,1);
+imshow(depth);
+colormap('parula');
+caxis([0,3000]);
+title('depth image');
+
+%% visualize mask
+mask = imread('mask.png');
+subplot(2,2,2);
+imshow(mask);
+title('segmentation mask');
+
+%% visualize skeleton from openpose
+% load openpose result
+fid = fopen('keypoint.json'); 
+raw = fread(fid,inf); 
+str = char(raw'); 
+fclose(fid); 
+val = jsondecode(str);
+pose2d = reshape(val.people.pose_keypoints_2d, 3, 25)';
+
+% visualize 2d result
+j_index = [9,10,11,12,13,14,15,20,21,22,23,24,25];
+j_tree = [9,10;10,11;11,12;12,23;12,25;23,24; 
+          9,13;13,14;14,15;15,20;15,22;20,21];
+subplot(2,2,3);
+plot(pose2d(j_index,1), pose2d(j_index,2),'.','MarkerSize',25);
+hold on;
+
+for j = 1:size(j_tree,1)
+    plot(pose2d(j_tree(j,:),1), pose2d(j_tree(j,:),2),'-', 'LineWidth', 3);    
+end
+
+imshow([]);
+xlim([400, 1000]);
+ylim([50, 600]);
+title('2D Keypoint');
+
+%% visualize point cloud
+load('pointcloud.mat');
+subplot(2,2,4);
+pcshow(pc_2048,'VerticalAxis','Y','VerticalAxisDir','down')
+hold on;
+pcshow(keypoint,'VerticalAxis','Y','VerticalAxisDir','down', 'MarkerSize', 300)
+title('\color{black}Point Cloud');
+set(gca, 'color', 'w');
+set(gcf, 'color', 'w');
+axis off;
+
+% visualize the body_orientation (of pelvis)
+plotTransforms((keypoint(1,:)+keypoint(5,:))/2, body_orientation, 'FrameSize',0.2);