dpernes / FCN-rLSTM

FCN-rLSTM for vehicle counting in city cameras (unofficial implementation)
https://arxiv.org/abs/1707.09476
13 stars 3 forks source link

Will there be a detailed instuction on how to implement the code #1

Closed Xinxinatg closed 3 years ago

Xinxinatg commented 3 years ago

I am trying to implement the code, whereas it's bit difficult to read the code and run it from scratch. Will there be any chance that the detailed instruction will be attached? Thanks in advance.

dpernes commented 3 years ago

What exactly do you want to implement?

Xinxinatg commented 3 years ago

Like how to import data, how to set up the hierarchy of folders? And how to set up the hyperparameters when training and testing? I am also thinking of applying this code on crowd counting dataset. Thanks a lot for your reply!

Xinxinatg commented 3 years ago

What exactly do you want to implement?

I think I figured it out almost. Just started learning deep learning 2 months ago, so bit confused when trying to run the codes. But just one quick question, the paper on which the code is based talked about this can also be implemented on UCSD pedestrian counting dataset, have you ever tried to implement this part on crowd counting as well?

dpernes commented 3 years ago

I think I figured it out almost.

Yeah, it should be pretty obvious from the arguments of train and test scripts. Just note that train_seq and test_seq do use sequential data (i.e. video) and therefore include the LSTM part. The versions without the suffix *_seq treat each frame separately, and therefore no LSTM is used.

this can also be implemented on UCSD pedestrian counting dataset, have you ever tried to implement this part on crowd counting as well?

Yes, I have. It's pretty straightforward to adapt the code for this dataset and it works nicely there too.

Xinxinatg commented 3 years ago

Thanks heaps!