MarStarck

why add accuraciesa with accuraciesb in eval?

1

in my opinion, the part a is used for update initial weight and the part b is used for inner learning. So in eval, part a can be used to...

when calculating loss, the CTCLoss method always returns 0, `cost = criterion(preds, text, preds_size, length) / batch_size` I've found out the `preds` size is `26*batch_size*label_size` the `text` size is total...

folly is compiled successfully, but problem occurs when make thrift: ``` /usr/local/lib/libfolly.so: undefined reference to `jump_fcontext' /usr/local/lib/libfolly.so: undefined reference to `make_fcontext' ```

detection_eval is high but demo_det.py find a lot of false positive boxes

3

my detection_eval is about 0.77~0.86 according to different iterations. but the detection results contains around 50% false positive boxes.

finetune doesn't converge

1

I trained the model using SVHN dataset and get around 80% accuracy on my own dataset. So I determine to finetune it on my dataset. ``` train_layers = ['hidden10', 'digit_length',...

Will it work for characters?

6

I wanna recognize images including short characters as well as digits, like "A25", "B03" and so on. I wonder could this structure work on this situation?

is there any difference for reinforcement learning implementation? maml has 2 repos, one for supervised and one for rl, I wonder why reptile has no rl version.

support for parsing fev file

4

i wanna parse information in fev file, any suggestions? thanks in advance

take_action in action.py has 28 types of actions, while in tools/action.py has only 12

what does the subdir mean

1

from "0000" to "1203"