Faster-rcnn.pytorch: Segmentation fault

[gufengnudt]

Hi, I have segmentation fault when I run demo.py in pytorch-1.0 branch. Now I only know the problem lies with NMS, but I don't know how to solve. Please help me, thanks a lot.

[winterest]

I have the same problem.

[yyq90]

same problem with cuda 10 cudnn 7 pytorch1.0 python 3.7

[Bijiazhou]

same problem with cuda 9.0 cudnn 7 pytorch1.0 python 3.7
Im sure abuot the problem with (demo.py line 300)

>       ```
> rois, cls_prob, bbox_pred, \
>       rpn_loss_cls, rpn_loss_box, \
>       RCNN_loss_cls, RCNN_loss_bbox, \
>       rois_label = fasterRCNN(im_data, im_info, gt_boxes, num_boxes)
> ```

[Bijiazhou]

same problem with cuda 9.0 cudnn 7 pytorch1.0 python 3.7
Im sure abuot the problem with (demo.py line 300)

>       ```
> rois, cls_prob, bbox_pred, \
>       rpn_loss_cls, rpn_loss_box, \
>       RCNN_loss_cls, RCNN_loss_bbox, \
>       rois_label = fasterRCNN(im_data, im_info, gt_boxes, num_boxes)
> ```

I fix it with chanage my env to py2.7 instead of py 3.7 ,in the master branch.
after that,i also try about py36,its also work.
so you'd better change to py27 or 36,and gcc version is also import,i use 5.5

[beckhamchen]

when I use python3.6, it also fails. But when change to python2.7, it works. gcc is updated to 5.2.0 (4.8.5 doesn't work)

[takahiro-itazuri]

I disable GPU version of NMS in line 196 in trainval_net.py", but it failed.
Please help me!

[takahiro-itazuri]

I won!!!

I checked and changed following things:

1. update GCC from version 4 to version 5 (I don't have root access, but "conda install -c cqwz gcc-5" and "conda install -c anaconda libgcc" are available!)
2. check my CUDA path correct.
3. move to pytorch-1.0 branch and run "python setup.py build develop"
4. clone coco and make.

The details of my environment are follows:
CUDA: 9.0
GCC: 5.2.0
LIBGCC: 7.2.0

[heekhero]

I tried all the above methods, but they didn't work.
The details of my environment are follows:
Python 3.6.8
CUDA 9.0
GCC 5.2.0
Pytorch 1.0.0
Ubuntu 14.04

Please help me!

[heekhero]

I won!!!

I checked and changed following things:

1. update GCC from version 4 to version 5 (I don't have root access, but "conda install -c cqwz gcc-5" and "conda install -c anaconda libgcc" are available!)
2. check my CUDA path correct.
3. move to pytorch-1.0 branch and run "python setup.py build develop"
4. clone coco and make.

The details of my environment are follows:
CUDA: 9.0
GCC: 5.2.0
LIBGCC: 7.2.0

so what is your version of python?

[WilyZhao8]

I met the same problem two days ago, I tried all the methods in these project, but I haven't solved it yet. The detail of my environment are follow:
Ubuntu:14.04
Pytorch:1.0.0
python:3.6
CUDA:8.0
GCC 5.2.0

[stiv-yakovenko]

I have the same problem.

[dhlee-aron]

I solved it. I just re-installed faster rcnn python2.7 pytorch 1.0.0.

I dont know why the error occurred but I guess
python 3.x make error during setup /lib/setup.py

you just use python 2.7, pytorch 1.0.0
update pycocotools and setup setup.py
enjoy!!

[Chida15]

I solved it. I just re-installed faster rcnn python2.7 pytorch 1.0.0.

I dont know why the error occurred but I guess
python 3.x make error during setup /lib/setup.py

you just use python 2.7, pytorch 1.0.0
update pycocotools and setup setup.py
enjoy!!

but when i use python2.7, there are some packages not work， such as pickle， how did you solve it?

[Chida15]

I solved it. I just re-installed faster rcnn python2.7 pytorch 1.0.0.

I dont know why the error occurred but I guess
python 3.x make error during setup /lib/setup.py

you just use python 2.7, pytorch 1.0.0
update pycocotools and setup setup.py
enjoy!!

and I still get segmentation fault