Closed anisghaoui closed 4 years ago
I think there is a conflict between the versions of torch/torchvision here.
$ python3 test_detection.py --split=sample --folder=sampledata Namespace(class_path='data/classes.names', conf_thres=0.5, folder='sampledata', img_size=608, model_def='config/complex_tiny_yolov3.cfg', nms_thres=0.5, split='sample', weights_path='checkpoints/tiny-yolov3_ckpt_epoch-220.pth') Load TEST samples from data/KITTI/object/sampledata Done: total TEST samples 12 Traceback (most recent call last): File "test_detection.py", line 124, in <module> detections = utils.non_max_suppression_rotated_bbox(detections, opt.conf_thres, opt.nms_thres) File "/home/anis/Complex-YOLOv3/utils/utils.py", line 302, in non_max_suppression_rotated_bbox invalid = large_overlap & label_match RuntimeError: _th_and not supported on CPUType for Bool
I replaced the & operator by np.logical_and() and it worked
&
np.logical_and()
This will require a fix
@anisghaoui Even, I got the same error. Thanks for the solution. Replacing the & operator by np.logical_and() worked!
invalid = np.logical_and(large_overlap,label_match)
I think there is a conflict between the versions of torch/torchvision here.
I replaced the
&
operator bynp.logical_and()
and it worked