fix reid recall metric bugs

ppcls/engine/trainer.py

@@ -442,10 +442,12 @@ class Trainer(object):
                     keep_mask = paddle.logical_or(query_id_mask, image_id_mask)
                     similarity_matrix = similarity_matrix * keep_mask.astype(
                         "float32")
+                else:
+                    keep_mask = None
 
                 metric_tmp = self.eval_metric_func(similarity_matrix,
                                                    image_id_blocks[block_idx],
-                                                   gallery_img_id)
+                                                   gallery_img_id, keep_mask)
 
                 for key in metric_tmp:
                     if key not in metric_dict:

ppcls/metric/metrics.py

@@ -16,6 +16,7 @@ import numpy as np
 import paddle
 import paddle.nn as nn
 
+
 class TopkAcc(nn.Layer):
     def __init__(self, topk=(1, 5)):
         super().__init__()
@@ -34,54 +35,72 @@ class TopkAcc(nn.Layer):
                 x, label, k=k)
         return metric_dict
 
+
 class mAP(nn.Layer):
     def __init__(self):
         super().__init__()
 
-    def forward(self, similarities_matrix, query_img_id, gallery_img_id):
+    def forward(self, similarities_matrix, query_img_id, gallery_img_id,
+                *args):
         metric_dict = dict()
-        
-        choosen_indices = paddle.argsort(similarities_matrix, axis=1, descending=True) 
-        gallery_labels_transpose = paddle.transpose(gallery_img_id, [1,0])
-        gallery_labels_transpose = paddle.broadcast_to(gallery_labels_transpose, shape=[choosen_indices.shape[0],  gallery_labels_transpose.shape[1]])
-        choosen_label = paddle.index_sample(gallery_labels_transpose, choosen_indices)        
+
+        choosen_indices = paddle.argsort(
+            similarities_matrix, axis=1, descending=True)
+        gallery_labels_transpose = paddle.transpose(gallery_img_id, [1, 0])
+        gallery_labels_transpose = paddle.broadcast_to(
+            gallery_labels_transpose,
+            shape=[
+                choosen_indices.shape[0], gallery_labels_transpose.shape[1]
+            ])
+        choosen_label = paddle.index_sample(gallery_labels_transpose,
+                                            choosen_indices)
         equal_flag = paddle.equal(choosen_label, query_img_id)
         equal_flag = paddle.cast(equal_flag, 'float32')
 
         acc_sum = paddle.cumsum(equal_flag, axis=1)
         div = paddle.arange(acc_sum.shape[1]).astype("float32") + 1
-        precision =  paddle.divide(acc_sum, div)
+        precision = paddle.divide(acc_sum, div)
 
         #calc map
         precision_mask = paddle.multiply(equal_flag, precision)
-        ap = paddle.sum(precision_mask, axis=1) / paddle.sum(equal_flag, axis=1)
+        ap = paddle.sum(precision_mask, axis=1) / paddle.sum(equal_flag,
+                                                             axis=1)
         metric_dict["mAP"] = paddle.mean(ap).numpy()[0]
         return metric_dict
 
+
 class mINP(nn.Layer):
     def __init__(self):
         super().__init__()
 
-    def forward(self, similarities_matrix, query_img_id, gallery_img_id):
+    def forward(self, similarities_matrix, query_img_id, gallery_img_id,
+                *args):
         metric_dict = dict()
-        
-        choosen_indices = paddle.argsort(similarities_matrix, axis=1, descending=True) 
-        gallery_labels_transpose = paddle.transpose(gallery_img_id, [1,0])
-        gallery_labels_transpose = paddle.broadcast_to(gallery_labels_transpose, shape=[choosen_indices.shape[0],  gallery_labels_transpose.shape[1]])
-        choosen_label = paddle.index_sample(gallery_labels_transpose, choosen_indices)        
+
+        choosen_indices = paddle.argsort(
+            similarities_matrix, axis=1, descending=True)
+        gallery_labels_transpose = paddle.transpose(gallery_img_id, [1, 0])
+        gallery_labels_transpose = paddle.broadcast_to(
+            gallery_labels_transpose,
+            shape=[
+                choosen_indices.shape[0], gallery_labels_transpose.shape[1]
+            ])
+        choosen_label = paddle.index_sample(gallery_labels_transpose,
+                                            choosen_indices)
         tmp = paddle.equal(choosen_label, query_img_id)
         tmp = paddle.cast(tmp, 'float64')
 
         #do accumulative sum
         div = paddle.arange(tmp.shape[1]).astype("float64") + 2
-        minus =  paddle.divide(tmp, div)
-        auxilary =  paddle.subtract(tmp, minus)
+        minus = paddle.divide(tmp, div)
+        auxilary = paddle.subtract(tmp, minus)
         hard_index = paddle.argmax(auxilary, axis=1).astype("float64")
         all_INP = paddle.divide(paddle.sum(tmp, axis=1), hard_index)
         mINP = paddle.mean(all_INP)
         metric_dict["mINP"] = mINP.numpy()[0]
         return metric_dict
 
+
 class Recallk(nn.Layer):
     def __init__(self, topk=(1, 5)):
         super().__init__()
@@ -90,25 +109,43 @@ class Recallk(nn.Layer):
             topk = [topk]
         self.topk = topk
 
-    def forward(self, similarities_matrix, query_img_id, gallery_img_id):
+    def forward(self, similarities_matrix, query_img_id, gallery_img_id,
+                keep_mask):
         metric_dict = dict()
 
         #get cmc
-        choosen_indices = paddle.argsort(similarities_matrix, axis=1, descending=True) 
-        gallery_labels_transpose = paddle.transpose(gallery_img_id, [1,0])
-        gallery_labels_transpose = paddle.broadcast_to(gallery_labels_transpose, shape=[choosen_indices.shape[0],  gallery_labels_transpose.shape[1]])
-        choosen_label = paddle.index_sample(gallery_labels_transpose, choosen_indices)        
+        choosen_indices = paddle.argsort(
+            similarities_matrix, axis=1, descending=True)
+        gallery_labels_transpose = paddle.transpose(gallery_img_id, [1, 0])
+        gallery_labels_transpose = paddle.broadcast_to(
+            gallery_labels_transpose,
+            shape=[
+                choosen_indices.shape[0], gallery_labels_transpose.shape[1]
+            ])
+        choosen_label = paddle.index_sample(gallery_labels_transpose,
+                                            choosen_indices)
         equal_flag = paddle.equal(choosen_label, query_img_id)
+        if keep_mask is not None:
+            keep_mask = paddle.index_sample(
+                keep_mask.astype('float32'), choosen_indices)
+            equal_flag = paddle.logical_and(equal_flag,
+                                            keep_mask.astype('bool'))
         equal_flag = paddle.cast(equal_flag, 'float32')
-        
+        real_query_num = paddle.sum(equal_flag, axis=1)
+        real_query_num = paddle.sum(
+            paddle.greater_than(real_query_num, paddle.to_tensor(0.)).astype(
+                "float32"))
+
         acc_sum = paddle.cumsum(equal_flag, axis=1)
-        mask = paddle.greater_than(acc_sum, paddle.to_tensor(0.)).astype("float32")
-        all_cmc = paddle.mean(mask, axis=0).numpy() 
+        mask = paddle.greater_than(acc_sum,
+                                   paddle.to_tensor(0.)).astype("float32")
+        all_cmc = (paddle.sum(mask, axis=0) / real_query_num).numpy()
 
         for k in self.topk:
             metric_dict["recall{}".format(k)] = all_cmc[k - 1]
         return metric_dict
 
+
 class DistillationTopkAcc(TopkAcc):
     def __init__(self, model_key, feature_key=None, topk=(1, 5)):
         super().__init__(topk=topk)
@@ -132,4 +169,3 @@ class GoogLeNetTopkAcc(TopkAcc):
 
     def forward(self, x, label):
         return super().forward(x[0], label)
-