error while tarining on my new dataset which has same COCO format

[alaa-shubbak]

I am trying to do training on my dataset which has the same format as the coco dataset , but with only one category (class) such class is one of the 80 classes in COCO datset.

I did the training on 2 GPU , so my batch_size= 2 and the chunk_size=[2] I only changed the chunk and the batch size in the config files of both CornerNet_Squeese and CornerNet_Saccade.

I did the training on both models , unfortunately, the training starts and stoped at some certain point with the following error message , the same error display while training the both models.

any help please will be with high benifites.

[alaa-shubbak]

Notice please that i did the change on the number and names of my new category in both the configuration json file and the model file (.py). so I change the number of classes from 80 to 1 in the json file change the same info in dbs> coco.py and change the number 80 into the focal loss in the model CornerNet_squeese ... but nothing change, I still face the same error.

i did the change as same as what has mentioned in this: https://github.com/huangzicheng/CornerNet-Lite

with more changing in the coco.py file