Init

README.md

@@ -1,2 +1,64 @@
 # 2s-AGCN
 Two-Stream Adaptive Graph Convolutional Networks for Skeleton-Based Action Recognition in CVPR19
+
+# Data Preparation
+
+ - Download the raw data from [NTU-RGB+D][https://github.com/shahroudy/NTURGB-D] and [Skeleton-Kinetics][https://github.com/yysijie/st-gcn]. Then put them under the data directory:
+
+        -data\  
+          -kinetics_raw\  
+            -kinetics_train\
+              ...
+            -kinetics_val\
+              ...
+            -kinetics_train_label.json
+            -keintics_val_label.json
+          -nturgbd_raw\  
+            -nturgb+d_skeletons\
+              ...
+            -samples_with_missing_skeletons.txt
+            
+
+[https://github.com/shahroudy/NTURGB-D]: NTU-RGB+D
+[https://github.com/yysijie/st-gcn]: Skeleton-Kinetics
+
+ - Preprocess the data with
+
+    `python data_gen/ntu_gendata.py`
+
+    `python data_gen/kinetics-gendata.py.`
+
+ - Generate the bone data with: 
+
+    `python data_gen/gen_bone_data.py`
+
+# Training & Testing
+
+Change the config file depending on what you want.
+
+
+    `python main.py --config ./config/nturgbd-cross-view/train_joint.yaml`
+
+    `python main.py --config ./config/nturgbd-cross-view/train_bone.yaml`
+To ensemble the results of joints and bones, run test firstly to generate the scores of the softmax layer. 
+
+    `python main.py --config ./config/nturgbd-cross-view/test_joint.yaml`
+
+    `python main.py --config ./config/nturgbd-cross-view/test_bone.yaml`
+
+Then combine the generated scores with: 
+
+    `python ensemble.py` --datasets ntu/xview
+     
+# Citation
+Please cite the following paper if you use this repository in your reseach.
+
+    @inproceedings{2sagcn2019cvpr,  
+      title     = {Two-Stream Adaptive Graph Convolutional Networks for Skeleton-Based Action Recognition},  
+      author    = {Lei Shi and Yifan Zhang and Jian Cheng and Hanqing Lu},  
+      booktitle = {CVPR},  
+      year      = {2019},  
+    }
+
+# Contact
+For any questions, feel free to contact: `lei.shi@nlpr.ia.ac.cn`

config/nturgbd-cross-subject/test_bone.yaml

@@ -17,10 +17,10 @@ model_args:
 
 # test
 phase: test
-device: [0,1,2,3]
+device: [4,5,6,7]
 test_batch_size: 256
-weights: ./pre_train/ntu_cs_agcn_bone-49-31300.pt
+weights: ./runs/ntu_cs_agcn_bone-49-31300.pt
 
-work_dir: ./runs/ntu/xsub/agcn_test_bone
+work_dir: ./work_dir/ntu/xsub/agcn_test_bone
 model_saved_name: ./runs/ntu_cs_agcn_test_bone
 save_score: True

config/nturgbd-cross-subject/train_bone.yaml

@@ -32,7 +32,7 @@ base_lr: 0.1
 step: [30, 40]
 
 # training
-device: [0, 1 ,2, 3]
+device: [4, 5, 6, 7]
 batch_size: 64
 test_batch_size: 64
 num_epoch: 50

ensemble.py

@@ -1,23 +1,28 @@
+import argparse
 import pickle
+
 import numpy as np
 from tqdm import tqdm
 
-# datasets = {'kinetics', 'ntu/xsub', 'ntu/xview'}
-datasets = {'ntu/xview'}
+parser = argparse.ArgumentParser()
+parser.add_argument('--datasets', default='ntu/xsub', choices={'kinetics', 'ntu/xsub', 'ntu/xview'},
+                    help='the work folder for storing results')
+parser.add_argument('--alpha', default=1, help='weighted summation')
+arg = parser.parse_args()
 
-for dataset in datasets:
-    label = open('./data/' + dataset + '/val_label.pkl', 'rb')
-    label = np.array(pickle.load(label))
-    r1 = open('./work_dir/' + dataset + '/agcn_test_joint/epoch1_test_score.pkl', 'rb')
-    r1 = list(pickle.load(r1).items())
-    r2 = open('./work_dir/' + dataset + '/agcn_test_bone/epoch1_test_score.pkl', 'rb')
-    r2 = list(pickle.load(r2).items())
+dataset = arg.datasets
+label = open('./data/' + dataset + '/val_label.pkl', 'rb')
+label = np.array(pickle.load(label))
+r1 = open('./work_dir/' + dataset + '/agcn_test_joint/epoch1_test_score.pkl', 'rb')
+r1 = list(pickle.load(r1).items())
+r2 = open('./work_dir/' + dataset + '/agcn_test_bone/epoch1_test_score.pkl', 'rb')
+r2 = list(pickle.load(r2).items())
 right_num = total_num = right_num_5 = 0
 for i in tqdm(range(len(label[0]))):
     _, l = label[:, i]
     _, r11 = r1[i]
     _, r22 = r2[i]
-    r = r11 + r22
+    r = r11 + r22 * arg.alpha
     rank_5 = r.argsort()[-5:]
     right_num_5 += int(int(l) in rank_5)
     r = np.argmax(r)

feeders/tools.py

@@ -1,6 +1,7 @@
-import numpy as np
 import random
 
+import numpy as np
+
 
 def downsample(data_numpy, step, random_sample=True):
     # input: C,T,V,M
@@ -50,7 +51,7 @@ def random_choose(data_numpy, size, auto_pad=True):
         else:
             return data_numpy
     else:
-        begin = random.randint(0, T - size)  # TODO
+        begin = random.randint(0, T - size)
         return data_numpy[:, begin:begin + size, :, :]
 
 
@@ -62,7 +63,7 @@ def random_move(data_numpy,
     # input: C,T,V,M
     C, T, V, M = data_numpy.shape
     move_time = random.choice(move_time_candidate)
-    node = np.arange(0, T, T * 1.0 / move_time).round().astype(int)  # TODO
+    node = np.arange(0, T, T * 1.0 / move_time).round().astype(int)
     node = np.append(node, T)
     num_node = len(node)