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

Low score in codalab with released codes. #27

Open xzjcs1992 opened 4 years ago

xzjcs1992 commented 4 years ago

Hello, Thank you for your interesting work and nice release of codes. I am trying to reproduce your results about VIS dataset. I successfully executed your codes and got the results.json file. However, when I upload the file to Codalab server, I got a very low score (0.2% AP). I can't determinate whether it's my environment problem (because there is no error alarm) or it's the problem of Codalab. Would you mind maybe provide your json file which achieves good score in codalab? Or maybe you have some better advices for me to diagnose my issue.

Attached please find my results.json results_vis_sipmask_ms_1x_final.zip

JialeCao001 commented 4 years ago

@xzjcs1992 How you generate the json file? Can you show your command?

xzjcs1992 commented 4 years ago

I use the command: python ./tools/test_video.py configs/sipmask/sipmask_r50_caffe_fpn_gn_1x.py ./work_dirs/vis_sipmask_ms_1x_final.pth --out results.pkl --eval segm --show --save_path= ./work_dirs/output_images

JialeCao001 commented 4 years ago

@xzjcs1992 Please use the test command, instead of saving image command: python ./tools/test_video.py configs/sipmask/sipmask_r50_caffe_fpn_gn_1x.py ./work_dirs/sipmask_r50_fpn_1x.pth --out results.pkl --eval segm

xzjcs1992 commented 4 years ago

It works. I thought the rescale is out of the model instance after the generation of json. It's actually inside the get_bboxes method. Thank you so much.

JialeCao001 commented 4 years ago

@xzjcs1992 Yes. I donot integrate this two functions together.