Error with info "argmax of an empty sequence" while training on COCO dataset
crazylyf opened this issue · comments
Hi all,
When I tried to train on COCO dataset, the program exit with information: ValueError: attempt to get argmax of an empty sequence. Has anyone encountered such an problem?
def train_net_multi_gpu(solver_prototxt, roidb, output_dir, pretrained_model, max_iter, gpus):
"""Train a Fast R-CNN network."""
roidb = filter_roidb(roidb)
uid = caffe.NCCL.new_uid()
Add the bold line to filter some images without any annotations.
thx @baiyancheng20