Pre-trained model about issue #22 about pedestron HOT 4 CLOSED

Yes, I think we can provide the pre-trained models, it will take some time. As far as your 8.5 vs ours 7.5 MR^-2 is concerned, I mean its in the same ballpark as reported in the paper. I would not too much worry about < =~1% of MR-^2 points. In anyways, I think we will try to provide the pre-trained models.

Secondly, can you share with us the converted annotations that you use for training Pedestron for Wider Person dataset ? Perhaps upload a google link or you can directly email at [email protected]

from pedestron.

Thank you and looking forward to your pre-trained model!
Maybe my explanation misunderstood you. I mean, the model I pre-trained on wider pedestrian by myself can get a little better result(about 0.3 MR^-2) than the model you provided in the readme.md(the part Benchmark→Pre-Trained models→ Cascade Mask R-CNN→6. WIDER PERSON and WIDER Pedestrian). I guess it is the model pre-trained on WIDER PERSON and WIDER Pedestrian from the subtitle.
Additionally, I am also interested in WIDER PERSON, and if I get the the converted annotations, I will sent it to you by email.

from pedestron.

I see, let me clarify the model in the README.md is only trained on WiderPedestrian. I think the description is misleading over there.

from pedestron.

Thanks, I get it. I am looking forward your pre-trained model. Good luck with your study!

from pedestron.