search

youtubevos / MaskTrackRCNN

MaskTrackRCNN for video instance segmentation based on mmdetection
Apache License 2.0
431 stars 75 forks source link

mAP of MaskTrackRCNN tested on codalab server #11

Closed Jiangfeng-Xiong closed 5 years ago

Jiangfeng-Xiong commented 5 years ago

I use the pretrained model epoch_12.pth from Google Drive, and then evaluate the results on the codalab server, but i got mAP: 0.248, however, it should be 0.3 as reported in the paper. Any suggestion? Thank you!

Jiangfeng-Xiong commented 5 years ago

And besides, when I load the model weights epoch_12.th, I got this message

unexpected key in source state_dict: track_head.fc.weight, track_head.fc.bias missing keys in source state_dict: track_head.fcs.1.bias, track_head.fcs.0.bias, track_head.fcs.0.weight, track_head.fcs.1.weight

linjieyangsc commented 5 years ago

@Jiangfeng-Xiong It seems you are using a wrong config file. Please make sure to use this file to load the model: configs/masktrack_rcnn_r50_fpn_1x_youtubevos.py

Jiangfeng-Xiong commented 5 years ago

屏幕快照 2019-07-31 上午10 09 36 @linjieyangsc

ZENGXH commented 5 years ago

@linjieyangsc I also got 24.8..

linjieyangsc commented 5 years ago

@Jiangfeng-Xiong @ZENGXH It seems that the previous checkpoint was problematic. I have updated it with the correct one. Please try again and let me know if any questions. Sorry for the inconvenience.

Jiangfeng-Xiong commented 5 years ago

it works now, thank you @linjieyangsc

voidtaoist commented 5 years ago

I use the pretrained model epoch_12.pth from Google Drive, and then evaluate the results on the codalab server, but i got mAP: 0.301, however, it should be 0.303 as reported in the paper. Any suggestion? Thank you!