search

rbgirshick / py-faster-rcnn

Faster R-CNN (Python implementation) -- see https://github.com/ShaoqingRen/faster_rcnn for the official MATLAB version
Other
8.09k stars 4.11k forks source link

Issue while training VGG16 #87

Open athus1990 opened 8 years ago

athus1990 commented 8 years ago

After trying the solution #13 Training my net throws this error: File "/usr/local/PYFASTER2/py-faster-rcnn/tools/../lib/roi_data_layer/minibatch.py", line 126, in _sample_rois labels[fg_rois_per_this_image:] = 0 IndexError: failed to coerce slice entry of type numpy.float64 to integer

what is causing this issue?

austinreiter commented 8 years ago

I've seen this with fast-rcnn but the issue is the same. I believe it's caused by more recent versions of numpy (>= 1.11 if I remember correctly?). The fix is to cast fg_rois_per_this_image to int. There are a few places in minibatch.py that require such a fix.

ricepot100 commented 8 years ago

Maybe your cuda and cudnn version is not very match. I use cuda 7.5 , cudnn v4, it could create VGG16_faster_rcnn_final.caffemodel successfully under output/default/voc_2007_trainval/

riadhayachi commented 7 years ago

i get this error when trying to train on voc2007 riadh@riadh-Veriton:~/py-faster-rcnn$ ./experiments/scripts/faster_rcnn_end2end.sh 0 VGG_CNN_M_1024 --set -e