These are chat archives for beniz/deepdetect

14th
Aug 2016
Jacky Yang
@anguoyang
Aug 14 2016 00:13
@sunatthegilddotcom , thank you so much!
sunatthegilddotcom
@sunatthegilddotcom
Aug 14 2016 03:53
@anguoyang you are welcome
Jacky Yang
@anguoyang
Aug 14 2016 10:00
File "imgsearch.py", line 78, in <module>
for p in classif ['body']['predictions']:
KeyError: 'body'
Emmanuel Benazera
@beniz
Aug 14 2016 10:54
there's no error handling in imgsearch, it's up to users. When a call fails, it fails.
you can PR error control typically
Jacky Yang
@anguoyang
Aug 14 2016 12:55
I find this: beniz/deepdetect#135
sunatthegilddotcom
@sunatthegilddotcom
Aug 14 2016 15:40
@beniz Hi. I think it might be easy for me to modle your building model. Is it possible for you to share the dataset so I can duplicate your work first? Thanks
Emmanuel Benazera
@beniz
Aug 14 2016 16:22
modle ? we don't share datasets due to copyright issues. You can build your own from Imagenet.
sunatthegilddotcom
@sunatthegilddotcom
Aug 14 2016 17:19
Thanks
sunatthegilddotcom
@sunatthegilddotcom
Aug 14 2016 17:33
@beniz I need help to organize the classification dataset, it has "166 convent" for instance, how should I use modify your building model to build the training dataset and then make any adjustment to your model to fit the new dataset.
Emmanuel Benazera
@beniz
Aug 14 2016 17:37
have you looked at the finetuning example ? and before that look at the tutorials to get to train your own model. It explains how to structure your dataset and it is very simple.
sunatthegilddotcom
@sunatthegilddotcom
Aug 14 2016 17:40
I try to make the connection between the different classification with the training dataset folder structure. In the google net corresp.txt, It has "20 n01601694 water ouzel, dipper", the dataset is organized by foldername "n01601694". is this the correct approach to link the classification with training dataset? Thank you very much
@beniz Thank you very much for the guidance. I have not looked the finetuning example yet. I am going to look at now. I am trying to follow all of the tutorial first.