Is end-to-end training able? about fcis HOT 8 CLOSED

Hi, the tech report is available at https://arxiv.org/abs/1611.07709

from fcis.

from fcis.

Hi,
Is there any ETA on the code release? I am working on modifying DeepMask now for the simple task of auto-cropping, and am pretty unhappy with how slow it is to train and the quality of the results. Would be nice to have the code of this to work with too since I only have a few more weeks to experiment...
-Thanks

from fcis.

Hi, the arxiv paper would be released quite soon within one week. We are working hard on coding releasing as well, but it would take some time. I cannot promise, but hopefully we can get it done in one month. You can also modify based on R-FCN (https://github.com/daijifeng001/R-FCN).

from fcis.

I see, thanks.

from fcis.

Thanks a lot.

from fcis.

Hi,
Can you tell me the threshold for predicting final results (vis_thresh) ? Is it 0.8 as in R-FCN or 0.5 as in MNC?
Is mask size = 14 as in Inst-FCN?

from fcis.

＠pqviet Hi, we use mask_size=21 in FCIS and vis_thresh=0.7 is used to visualize the result.

from fcis.