train error!!! about centernet HOT 9 OPEN

@Duankaiwen

from centernet.

this error the same to me
please tell me how to resolve it
thank you

from centernet.

what's your torch version and gcc version?

from centernet.

@Duankaiwen
torch is 0.4.1

from centernet.

Does the original code work in your own computer?

from centernet.

no , because i had download the coco images before and the image is too big, so i use the images which i download before and use your json file . ,but have the error

your train2014 json file have the val2014 image?

from centernet.

@Duankaiwen

from centernet.

Try using the gcc 4.9 to recompile the corner pooling layers.

from centernet.

@Duankaiwen Thanks for the amazing work! I have been able to run your code as instructed on the COCO dataset and run some experiments on it successfully. However on my own dataset (formatted in COCO style), it starts the training, however it just stops abruptly, with no error or warning. I'm unable to debug this. gcc version is 5.4.0, Ubuntu 18.04 and PyTorch is 1.0.0. Here's what the screen looks like:

from centernet.