ValueError: attempt to get argmax of an empty sequence-蒲公英云

ValueError: attempt to get argmax of an empty sequence

墨蓝 2023-06-21 10:23 126阅读 0赞

本文针对faster rcnn训练中出现上述问题及解决的方法进行简单记录。

报错信息

2019-12-12 04:51:57.002253: W tensorflow/core/framework/op_kernel.cc:1610] Invalid argument: ValueError: attempt to get argmax of an empty sequence
Traceback (most recent call last):

File “/usr/local/lib/python3.5/dist-packages/tensorflowcore/python/ops/scriptops.py”, line 221, in __call
```
ret = func(*args)
```
File “/workspace/tf-faster-rcnn-master/tools/../lib/layer_utils/anchor_target_layer.py”, line 51, in anchor_target_layer
```
argmax_overlaps = overlaps.argmax(axis=1)
```
排查方法
1. 增加文件名信息至训练数据(lib/roi_data_layer/minibatch.py程序的get_minibatch函数)
  blobs[‘addimages’]=roidb[0][‘image’] #增加语句
  blobs[‘gt_boxes’] = gt_boxes
2. 输出当前训练batch数据的信息(lib/model/train_val.py程序的construct_graph函数)
  blobs = self.data_layer.forward()
  print (‘iter{}, info:{}’.format(iter,blobs)) #增加语句
3. 输出当前val batch数据的信息(lib/model/train_val.py程序的construct_graph函数)
  blobs_val = self.data_layer_val.forward()
  print (‘iter{}, info_val:{}’.format(iter,blobs)) #增加语句

重新运行获得错误数据的图片信息

iter4559, info_val:{‘addimages’: ‘data/JPEGImages/000202.jpg’, ‘data’: array([[[[ -91.9801 , -99.9465 , -107.7717 ],

     [ -91.5426    ,  -99.509     , -107.3342    ],
     ...,
     [  -6.9801    ,   25.0535    ,    1.2283    ]]]], dtype=float32), 'gt_boxes': array([], shape=(0, 5), dtype=float32), 'im_info': array([600. , 800. ,   1.6], dtype=float32)}

解决方法

查看xml标注文件，删除空元素
对于修改文件处理程序的，如pascal_voc.py，需要确保数据读取中无异常\

ps：可通过anchor_target_layer.py函数快速输出gt_boxes的信息

print ('gt_boxes: {}, ahcnors: {}'.format(gt_boxes, anchors))  #添加语句
  overlaps = bbox_overlaps(
    np.ascontiguousarray(anchors, dtype=np.float),
    np.ascontiguousarray(gt_boxes, dtype=np.float))
  argmax_overlaps = overlaps.argmax(axis=1)