Where is the config file sipmask_r50_caffe_fpn_gn_1x_4gpu.py of sipmask-vis?

[zack624]

Hello, thanks for your good work. I cannot find the config file sipmask_r50_caffe_fpn_gn_1x_4gpu.py in the vis project. There is just the configs/sipmask/sipmask_r50_caffe_fpn_gn_1x.py, and I have train with it, but got near 28 mAP in the evaluation website. So I want to ask what should I do more to get higher mAP. Thanks.

[JialeCao001]

@zack624 Thanks for interests. how many gpus are you using for training?

[zack624]

@zack624 Thanks for interests. how many gpus are you using for training?

Just one. I run directly "python tools/train.py configs/sipmask/sipmask_r50_caffe_fpn_gn_1x.py". And I also download your checkpoints to infer, but there is a lot of missing keys like that:

`
unexpected key in source state_dict: bbox_head.cls_convs.2.conv.weight, bbox_head.cls_convs.2.gn.weight, bbox_head.cls_convs.2.gn.bias, bbox_head.reg_convs.3.conv.weight, bbox_head.reg_convs.3.gn.weight, bbox_head.reg_convs.3.gn.bias, bbox_head.track_convs.2.conv.weight, bbox_head.track_convs.2.gn.weight, bbox_head.track_convs.2.gn.bias, bbox_head.track_convs.3.conv.weight, bbox_head.track_convs.3.gn.weight, bbox_head.track_convs.3.gn.bias, bbox_head.fovea_track.weight, bbox_head.fovea_track.bias

missing keys in source state_dict: bbox_head.sipmask_track.bias, bbox_head.sipmask_track.weight
`

A lot of input pictures are not be inferred, and the output file of json is just 200K.

[JialeCao001]

@zack624 you means that the provided inference model and the config file in vis donot match?

[zack624]

@JialeCao001 Thanks for your answer. I guess there might be some differences between the provided model file with the sipmask_r50_caffe_fpn_gn_1x.py config, so I ask for the sipmask_r50_caffe_fpn_gn_1x_4gpu.py mentioned in project readme.

[JialeCao001]

@zack624 Hi. I remove the _gpu from file name. I test on my device. It seems no error.

[PanXiongAdam]

Hi, I have the same problem about the near 28 mAP in the evaluation website. Have you sloved it, please? If sloved it, can you tell me how it do? Thank you very much.