search

JialeCao001 / SipMask

SipMask: Spatial Information Preservation for Fast Image and Video Instance Segmentation (ECCV2020)
https://arxiv.org/pdf/2007.14772.pdf
MIT License
334 stars 54 forks source link

Sipmask-VIS :output_size of sipmask_r50_caffe_1x.pth do not match #31

Closed YYlvsy closed 3 years ago

YYlvsy commented 3 years ago

First thanks for providing the code of this wonderful work! Here's the problem. When loading from './../SipMask-mmdetection/work_dirs/sipmask_r50_caffe_1x.pth', the output size of this pretrained model do not match it of the VIS model.Since the number of classes of the pretrained model is 80,different from it (which is 40) of the VIS model. Please tell me how to deal with this problem. Thank you!!

JialeCao001 commented 3 years ago

@YYIvsy Hi. Thanks for your interest. In fact, it is different. Does it report the error? or Just a warning? If it reports error, maybe you need to change the version of mmcv. Please use mmcv0.2.12.

YYlvsy commented 3 years ago

@YYIvsy Hi. Thanks for your interest. In fact, it is different. Does it report the error? or Just a warning? If it reports error, maybe you need to change the version of mmcv. Please use mmcv0.2.12.

Thanks a lot! I checked the version of mmcv,it turned out that I was using mmcv 0.2.10.I've already changed the version .Hope it works.