py-faster-rcnn 训练常见错误

在训练 py-faster-rcnn 时，常会出现以下错误：

• TypeError: ‘numpy.float64’ object cannot be interpreted as an index
• Check failed: error == cudaSuccess (4 vs. 0) unspecified launch failure
• AttributeError: ‘module’ object has no attribute ‘text_format’

---

TypeError: ‘numpy.float64’ object cannot be interpreted as an index

• 解决方法 sudo pip install -U numpy==1.11.0`

Check failed: error == cudaSuccess (4 vs. 0) unspecified launch failure

• 解决方法
  
  • ① 安装cudnn：

            cp caffe/inlude/caffe/layers/cudnn_* caffe-fast-rcnn/include/caffe/layers/            cp caffe/src/caffe/layers/cudnn_* caffe-fast-rcnn/src/caffe/layers/            cp caffe/include/caffe/util/cudnn.hpp caffe-fast-rcnn/include/caffe/util/            cd caffe-fast-rcnn            make clean            make -j8            make pycaffe

 - ② 修改配置文件 py-faster-rcnn/lib/fast_rcnn/config.py：将BATCH_SIZE调小，比如64,32,16,8....

        # Minibatch size (number of regions of interest [ROIs])        __C.TRAIN.BATCH_SIZE = 128

AttributeError: ‘module’ object has no attribute ‘text_format’

• 在py-faster-rcnn/lib/fast_rcnn/train.py中添加： import google.protobuf.text_format