search

ChenYingpeng / caffe-yolov3

A real-time object detection framework of Yolov3/v4 based on caffe
473 stars 231 forks source link

out of memory #24

Open MRRRKING opened 5 years ago

MRRRKING commented 5 years ago

首先，非常感谢您能分享出这么好的代码。 下面我说一下我遇到的问题，编译好您的代码之后，在运行detectnet时，出现了下面的问题：

num_inputs is 1 num_outputs is 3 I0117 16:22:17.890136 7075 detectnet.cpp:75] Input data layer channels is 3 I0117 16:22:17.890156 7075 detectnet.cpp:76] Input data layer width is 608 I0117 16:22:17.890162 7075 detectnet.cpp:77] Input data layer height is 608 Cannot load image "/home/jincan/deeplearning/examples/images/cat.jpg" F0117 16:22:18.198128 7075 syncedmem.cpp:71] Check failed: error == cudaSuccess (2 vs. 0) out of memory Check failure stack trace: 已放弃 (核心已转储)

我知道这是显存不足导致的（我用的是GTX 950M，2G显存），但是有没有什么方法可以解决这个问题，除了换显卡？ 非常感谢！

ChenYingpeng commented 5 years ago

你可以把输入尺寸变小一点，比如416x416，320x320等等，这样你消耗的显存会变小，也许就不会出现这个错误了。

MRRRKING commented 5 years ago

非常感谢，我把输入尺寸调到320*320时，便可以运行了。但是新的问题出现了，物体检测的位置不正确。 我刚开始使用的是您的yolov3_darknet2caffe.py的代码，在看了https://github.com/ChenYingpeng/caffe-yolov3/issues/11 这个问题贴之后，我又用了这个帖子https://github.com/ChenYingpeng/caffe-yolov3/issues/1 中您给的代码 ，但是，还是不能得到正确的输出。请问我应该怎么办？非常感谢。 2019-01-21 14-02-29

jackgao0323 commented 5 years ago

I remember the reason of this problem is it has set the output size of 416*416 in yolo_layer.cpp.

MRRRKING commented 5 years ago

Thanks a lot. I have solved it.

MRRRKING commented 4 years ago

Thanks a lot. I have solved it.

have you done with the position problem?

I didn't understand your question. Could you tell me in detail, please?