support stgcn trainer when classes < 5

Author: zoooo0820

paddlevideo/metrics/skeleton_metric.py

@@ -31,26 +31,29 @@ class SkeletonMetric(BaseMetric):
     Args:
         out_file: str, file to save test results.
     """
+
     def __init__(self,
                  data_size,
                  batch_size,
                  out_file='submission.csv',
-                 log_interval=1):
+                 log_interval=1,
+                 top_k=5):
         """prepare for metrics
         """
         super().__init__(data_size, batch_size, log_interval)
         self.top1 = []
         self.top5 = []
         self.values = []
         self.out_file = out_file
+        self.k = top_k
 
     def update(self, batch_id, data, outputs):
         """update metrics during each iter
         """
         if len(data) == 2:  # data with label
             labels = data[1]
             top1 = paddle.metric.accuracy(input=outputs, label=labels, k=1)
-            top5 = paddle.metric.accuracy(input=outputs, label=labels, k=2)
+            top5 = paddle.metric.accuracy(input=outputs, label=labels, k=self.k)
             if self.world_size > 1:
                 top1 = paddle.distributed.all_reduce(
                     top1, op=paddle.distributed.ReduceOp.SUM) / self.world_size

paddlevideo/modeling/backbones/stgcn.py

@@ -68,6 +68,7 @@ def normalize_digraph(A):
 
 
 class Graph():
+
     def __init__(self,
                  layout='openpose',
                  strategy='uniform',
@@ -109,16 +110,15 @@ def get_edge(self, layout):
             neighbor_link = [(i - 1, j - 1) for (i, j) in neighbor_1base]
             self.edge = self_link + neighbor_link
             self.center = 21 - 1
-        elif layout == 'ntu-rgb+d_fall':
+        elif layout == 'coco_keypoint':
             self.num_node = 17
             self_link = [(i, i) for i in range(self.num_node)]
-            neighbor_1base = [(0, 1), (0, 2), (1, 3), (2, 4), (3, 5), (4, 6), (5, 7), (6, 8),
-                 (7, 9), (8, 10), (5, 11), (6, 12), (11, 13), (12, 14),
-                 (13, 15), (14, 16), (11, 12)]
+            neighbor_1base = [(0, 1), (0, 2), (1, 3), (2, 4), (3, 5), (4, 6),
+                              (5, 7), (6, 8), (7, 9), (8, 10), (5, 11), (6, 12),
+                              (11, 13), (12, 14), (13, 15), (14, 16), (11, 12)]
             neighbor_link = [(i, j) for (i, j) in neighbor_1base]
             self.edge = self_link + neighbor_link
             self.center = 11
-
         else:
             raise ValueError("Do Not Exist This Layout.")
 
@@ -158,6 +158,7 @@ def get_adjacency(self, strategy):
 
 
 class ConvTemporalGraphical(nn.Layer):
+
     def __init__(self,
                  in_channels,
                  out_channels,
@@ -188,6 +189,7 @@ def forward(self, x, A):
 
 
 class st_gcn_block(nn.Layer):
+
     def __init__(self,
                  in_channels,
                  out_channels,
@@ -252,6 +254,7 @@ class STGCN(nn.Layer):
         edge_importance_weighting: bool, whether to use edge attention. Default True.
         data_bn: bool, whether to use data BatchNorm. Default True.
     """
+
     def __init__(self,
                  in_channels=2,
                  edge_importance_weighting=True,

paddlevideo/modeling/framework/recognizers/recognizer_gcn.py

@@ -21,6 +21,21 @@
 class RecognizerGCN(BaseRecognizer):
     """GCN Recognizer model framework.
     """
+
+    def __init__(self,
+                 backbone=None,
+                 head=None,
+                 runtime_cfg=None,
+                 if_top5=True):
+        """
+        Args:
+            backbone (dict): Backbone modules to extract feature.
+            head (dict): Classification head to process feature.
+            is_top5 (bool): Whether to display top-5 accuracy during training/validation steps.
+        """
+        super(RecognizerGCN, self).__init__(backbone, head, runtime_cfg)
+        self.if_top5 = if_top5
+
     def forward_net(self, data):
         """Define how the model is going to run, from input to output.
         """
@@ -36,7 +51,7 @@ def train_step(self, data_batch):
 
         # call forward
         cls_score = self.forward_net(data)
-        loss_metrics = self.head.loss(cls_score, label)
+        loss_metrics = self.head.loss(cls_score, label, if_top5=self.if_top5)
         return loss_metrics
 
     def val_step(self, data_batch):
@@ -47,7 +62,10 @@ def val_step(self, data_batch):
 
         # call forward
         cls_score = self.forward_net(data)
-        loss_metrics = self.head.loss(cls_score, label, valid_mode=True)
+        loss_metrics = self.head.loss(cls_score,
+                                      label,
+                                      valid_mode=True,
+                                      if_top5=self.if_top5)
         return loss_metrics
 
     def test_step(self, data_batch):

paddlevideo/modeling/heads/base.py

@@ -155,7 +155,7 @@ def label_smooth_loss(self, scores, labels, **kwargs):
     def get_acc(self, scores, labels, valid_mode, if_top5=True):
         if if_top5:
             top1 = paddle.metric.accuracy(input=scores, label=labels, k=1)
-            top5 = paddle.metric.accuracy(input=scores, label=labels, k=2)
+            top5 = paddle.metric.accuracy(input=scores, label=labels, k=5)
             _, world_size = get_dist_info()
             #NOTE(shipping): deal with multi cards validate
             if world_size > 1 and valid_mode:  #reduce sum when valid

paddlevideo/tasks/test.py

@@ -84,5 +84,4 @@ def test_model(cfg, weights, parallel=True):
         else:
             outputs = model(data, mode='test')
             Metric.update(batch_id, data, outputs)
-            print(outputs)
     Metric.accumulate()