search

Bartzi / see

Code for the AAAI 2018 publication "SEE: Towards Semi-Supervised End-to-End Scene Text Recognition"
GNU General Public License v3.0
575 stars 147 forks source link

train_fsns.py problem #71

Open MS-MA opened 5 years ago

MS-MA commented 5 years ago

In train_fsns.py, to trainer.run(), where does the program jump?

Bartzi commented 5 years ago

it jumps into the train routine implemented by chainer

MS-MA commented 5 years ago

Can you tell me the exact location of the program jump? thank you very much @Bartzi

Bartzi commented 5 years ago

please have a look at the chainer documentation, and I think you will be able to find what you are looking for :wink:

MS-MA commented 5 years ago

hello Bartzi,I have a new problem about train_fsns.py if i create a learning curriculum,like the follows:

[ { "train":"/home/data/fsns/image/train/train_swap_max2words.csv", "validation":"/home/data/fsns/image/validation/validation_swap_max2words.csv" }, { "train":"/home/data/fsns/image/train/train_swap_3words.csv", "validation":"/home/data/fsns/image/validation/validation_swap_3words.csv" }, { "train":"/home/data/fsns/image/train/train_swap_4words.csv", "validation":"/home/data/fsns/image/validation/validation_swap_4words.csv" }, { "train":"/home/data/fsns/image/train/train_swap_5words.csv", "validation":"/home/data/fsns/image/validation/validation_swap_5words.csv" }, { "train":"/home/data/fsns/image/train/train_swap_6words.csv", "validation":"/home/data/fsns/image/validation/validation_swap_6words.csv" }

]

how should i adjust the attributes like the below in train_fsns.py ?

attributes_to_adjust = [ ('num_timesteps', ['predictor', 'localization_net']), ('num_timesteps', ['predictor', 'recognition_net']), ('num_timesteps', ['lossfun', 'self']), ('num_labels', ['predictor', 'recognition_net']), ]

I am confused,could you give me a case? very thanks

Bartzi commented 5 years ago

If you are not changing anything else than a train curriculum there is no need for chaning the code at this point. These lines of code are only necessary, because there are sveral pieces of code that need to be adjusted if the dataset is changed, this is just a list of locations where changes need to be made when updating the curriculum, so as long as you did not make any changes to code that deals with attributes that might be changed by a new dataset, you don't need to changes these lines of code.