Network training guidance and model

[zhanghanduo]

Hi. Thanks for the inspirational work!

Just want to ask could you directly provide a downloadable link of COCO stuff weight file? If you have time, could you also share some of your tips of training this network like learning rate and stopping conditions because they are not explained in detail in the paper. Also if I want the function of "resuming latest epoch of model", I have to modify the code right?
Thanks again.

[nmerrill67]

Hi thanks for the interest. Did you mean the network weights or the loss weights? If you mean the loss weights, they are available in dataset/loss_weights.txt. I can provide network weights as well if that's what you meant.

As for training, I am using the parameters listed in the code (default Adam parameters). I stopped the model when is did not get any better testing on the CampusLoop dataset. This test is automatically run periodically during training as the eval functionality, and can be viewed from tensorboard under AUC.

[chuan573906361]

Hi Nathaniel, Thanks for the remarkable work!
I've trained the net on the coco-2017 dataset.then I tested the model on the campusLoopDataset provided in the repo.but it seems the performance is too much lower than the results in the paper.so could you please send me the model? [email protected]

[wtiandong]

Hi @nmerrill67 ,
Thanks for your great work! Could you send me the network weights to [email protected], please?

[dianqiliuze]

Hi Nathaniel, Thanks for your work! Could you please send me the network weights to [email protected]? thank you very much!

[garriton]

Hi Nathaniel, Thanks for your work! Could you please send me the network weights to [email protected]? thank you very much!

[chenjiangtao229]

Hi Nathaniel, Thanks for your work! Could you please send me the network weights to [email protected]? Thank you very much!

[zhuhu00]

Hi @nmerrill67 ,
Thanks for your great work! Could you send me the network weights to [email protected], please? Thank you very much.

[jialeB]

Hi, @nmerrill67 ，
Thanks for the inspirational work! Could you send me the network weights to [email protected], please? Thank you very much.

[dianqiliuze]

 您好，您的邮件我已收到，谢谢！

[xiaozhang404-coding]

Hi Nathaniel, Thanks for your work!I also trained the network at coco-2017, but after I tested the model, I found that the performance is lower than the results of the paper, can you send your model? My mailbox is [email protected]

[ZhangHaoHITSZ]

Hi Nathaniel, Thanks for your work! I want to use it in my own slam system but my GPU is so old that I can not train the net. So can you send your model and I would appreciate that. My mailbox is [email protected].

[dianqiliuze]

 您好，您的邮件我已收到，谢谢！

[dingshuoshuo]

Hi, @nmerrill67 ，
Thanks for the inspirational work! Could you send me the network weights to [email protected], please? Thank you very much.

[dianqiliuze]

 您好，您的邮件我已收到，谢谢！