can't load R50 FPN 2x checkpoint from Detectron

[happynewya]

Expected results

load checkpoints from Detectron R-50_FPN_2x

Actual results

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
~/work/Detectron.pytorch/tools/train_net_step.py in <module>()
    457
    458 if __name__ == '__main__':
--> 459     main()

~/work/Detectron.pytorch/tools/train_net_step.py in main()
    307     if args.load_detectron:  #TODO resume for detectron weights (load sgd momentum values)
    308         logging.info("loading Detectron weights %s", args.load_detectron)
--> 309         load_detectron_weight(maskRCNN, args.load_detectron)
    310
    311     lr = optimizer.param_groups[0]['lr']  # lr of non-bias parameters, for commmand line outputs.

~/work/Detectron.pytorch/lib/utils/detectron_weight_helper.py in load_detectron_weight(net, detectron_weight_file)
     21         d_name = name_mapping[p_name]
     22         if isinstance(d_name, str):  # maybe str, None or True
---> 23             p_tensor.copy_(torch.Tensor(src_blobs[d_name]))
     24
     25

KeyError: 'conv_rpn_fpn2_w'

Detailed steps to reproduce

python tools/train_net_step.py --dataset coco2017 --cfg configs/baselines/e2e_faster_rcnn_R-50-FPN_2x.yaml --bs 4 --load_detectron data/pretrained_model/detectron/e2e_faster_rcnn_R-50-FPN-2x.pkl

System information

• Operating system: Ubuntu
• CUDA version: 8.0
• python version: 3.5
• pytorch version: 0.4
• Anything else that seems relevant: I can load R-50-FPN-1x successfully，I think you can add a key check at load_detectron_weights。