search

HKUST-Aerial-Robotics / Stereo-RCNN

Code for 'Stereo R-CNN based 3D Object Detection for Autonomous Driving' (CVPR 2019)
MIT License
690 stars 177 forks source link

Segmentation fault #6

Closed Jokoe66 closed 5 years ago

Jokoe66 commented 5 years ago

when I run the test_net.py for evaluation with sh test.sh, I got following information :

Loaded dataset `kitti_val` for training
Preparing training data...
cache file ./data/kitti/object/kitti_val_gt_roidb.pkl
kitti_val gt roidb loaded from ./data/kitti/object/kitti_val_gt_roidb.pkl
done
3769 roidb entries
load checkpoint models_stereo/stereo_rcnn_18_6477.pth
load model successfully!
Segmentation fault (core dumped)s, solve time 0.41s (Press Esc to exit!)

It seems evaluation works well for a while, but after evalutating about 50 val samples, Segmentation fault is reported. Btw I use cuda8.0 and pytorch 0.3.1, and the training process works just fine. @PeiliangLi

Jokoe66 commented 5 years ago

seems it's working well with pytorch 0.3.0

PeiliangLi commented 5 years ago

OK, great