Training workflow does not work on Visual Genome

[YerongLi2]

python tools/train_net_step_rel.py --dataset vg8k --cfg configs/vg8k/e2e_relcnn_VGG16_8_epochs_vg8k_y_loss_only_hubness100k.yaml --nw 8 --use_tfboard --seed 3

INFO train_net_step_rel.py: 663: Save ckpt on exception ...
INFO train_net_step_rel.py: 150: save model: Outputs/e2e_relcnn_VGG16_8_epochs_vg8k_y_loss_only_hubness100k/vg8k/Apr02-00-17-05_bock_step_with_prd_cls
_v3/ckpt/model_step0.pth
INFO train_net_step_rel.py: 665: Save ckpt done.
Traceback (most recent call last):
  File "tools/train_net_step_rel.py", line 604, in main
    net_outputs = maskRCNN(**input_data)
  File "/home/wentaoy2/anaconda3/envs/vr/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/wentaoy2/LTVRR/lib/nn/parallel/data_parallel.py", line 108, in forward
    outputs = [self.module(*inputs[0], **kwargs[0])]
  File "/home/wentaoy2/anaconda3/envs/vr/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/wentaoy2/LTVRR/lib/modeling/model_builder_rel.py", line 513, in forward
    return self._forward(data, im_info, dataset_name, roidb, use_gt_labels, include_feat, **rpn_kwargs)
  File "/home/wentaoy2/LTVRR/lib/modeling/model_builder_rel.py", line 553, in _forward
    assert len(roidb) == 1

AssertionError

[aniket-agarwal1999]

Hi,
I am not sure why this could be occurring. I tried to reproduce this by a completely new setup but did not get this error. Can I once know if you have properly compiled the Detectron submodule along with all the other instructions on the README?