weirme / FCSN

A PyTorch reimplementation of FCSN in paper "Video Summarization Using Fully Convolutional Sequence Networks"
114 stars 33 forks source link

Pre-trained model #6

Open thisisjasleen opened 5 years ago

thisisjasleen commented 5 years ago

Could you please provide a pre-trained model for the same?

thisisjasleen commented 5 years ago

I'm trying to run the train.py file. However, I'm getting an error in this line: tqdm.write(f'\nEpoch {epoch_i}') Kindly help and please reply asap! Thanks in advance!

SinDongHwan commented 5 years ago

@thisisjasleen Hi, This is pretraind model(checkpoint)

Could you tell me error message? I want to run your code. Did you edit self.data_path in config.py? if not, edit self.data_path='./fcsn_dataset.h5' -> './fcsn_TVSum.h5'

thisisjasleen commented 5 years ago

@SinDongHwan Hi, thankyou for replying. I did edit the self.data_path . The dataset was downloaded under the name 'fcsn_TVSum.h5' but i changed it to the one already mentioned in the code i.e. 'fcsn_dataset.h5' The error still exists. A syntax error.

SinDongHwan commented 5 years ago

@thisisjasleen If you're okay, Can i see your code?? please send God.DongHwan@gmail.com or upload your repository.

thisisjasleen commented 5 years ago

@SinDongHwan I've used the same code as in this repo. Nothing has been changed.

thisisjasleen commented 5 years ago

@SinDongHwan Can you share your code with me? The complete code for the implementation of this paper. Here is my email: jasleenrissam8@gmail.com

SinDongHwan commented 5 years ago

@thisisjasleen Okay^^ I sent codes.

thisisjasleen commented 5 years ago

@SinDongHwan Is your code complete? Is there any additional thing that needs to be added or should i directly run it?

thisisjasleen commented 5 years ago

@SinDongHwan Also, I'm unable to open your pkl files.

SinDongHwan commented 5 years ago

@thisisjasleen

I just edited "self.data_path" config.py. and then execute. python train.py. i didn't face error.

if you face error, please error message, where line, python version. I can help exactly, because i don't know your situation detail.

thisisjasleen commented 5 years ago

@SinDongHwan Could the error be due to tqdm? I tried to run train.py in the exact same manner without changing anything in any of the codes bt still faced an error.

SinDongHwan commented 5 years ago

@thisisjasleen I think so, if just i see your statement. i don't know detail error. could you tell me your error message?

thisisjasleen commented 5 years ago

File train.py, Line 74 tqdm.write(f'\nEpoch {epoch_i}') Syntax Error:invalid syntax

SinDongHwan commented 5 years ago

what is your python version??

thisisjasleen commented 5 years ago

@SinDongHwan I've installed tqdm. Don't know why is this error still coming.

thisisjasleen commented 5 years ago

Python 3.5.2

SinDongHwan commented 5 years ago

Can you edit Line 74? tqdm.write('\nEpoch {}'.format(epoch_i)) please tell me, whether error on not. if not, edit tqdm.write code like this using.format()

thisisjasleen commented 5 years ago

Thanks. That issue seems resolved now. However, now I'm encountering a syntax error in this line: ckpt_path = self.config.save_dir + f'/epoch-{epoch_i}.pkl'

SinDongHwan commented 5 years ago

Good!! I think that "f" is string format. this code seems to support python 3.6 over. python3.6 don't face error. Please edit f'{epoch_i}' like this '{}'.format(epoch_i) i will edit code. and then pull requests.

thisisjasleen commented 5 years ago

Okay. Thankyou for the help. Also, have you sent me the codes? Are those codes complete or do i have to add things?

SinDongHwan commented 5 years ago

@thisisjasleen Yes. But just this repository codes. so you can't. Do you want executable codes?? I think nothing for train and test. maybe... hah

thisisjasleen commented 5 years ago

@SinDongHwan Yes, I need complete executable codes.Preferably the trained models which i can further tune as per my needs. Could you provide them?

thisisjasleen commented 5 years ago

ckpt_path = self.config.save_dir + f'/epoch-{epoch_i}.pkl' I tried editing this line like the way you suggested. It's showing some error though.

SinDongHwan commented 5 years ago

I just have a model made by this repository. you can get same model, if you just execute. f'/epoch-{epoch_i}.pkl' -> '/epoch-{}.pkl'.format(epoch_i) It's easy^^.

thisisjasleen commented 5 years ago

Can you send me your model? Is it the one you sent previously? I'm not able to open those pickle files.

SinDongHwan commented 5 years ago

You means that pickle files crushed? or can't download?

thisisjasleen commented 5 years ago

I've downloaded them. They just aren't opening.

thisisjasleen commented 5 years ago

@SinDongHwan I'm having a little trouble visualizing what the output of this entire code is>? I mean, what is the end result? Are we getting the summarized versions of the videos or not?

SinDongHwan commented 5 years ago

@thisisjasleen I think it's not hard to visualize. this result is sequence of {0,1}. if you see, 'score_dir/epoch-{}.json',

key is video number.
value is sequence of {0,1}. 1: keyframe, 0:not

you should post process. features of dataset pick one frame per 15 frame. and then this is used to train, evaluate, test.

to get summarize video, read #3 . so you can see evaluate code link.

thisisjasleen commented 5 years ago

Okay, I'll check the link. Btw, what does this pickle file contain?

Ai-is-light commented 3 years ago

pretraind model(checkpoint)

would you mind sharing your pre-trained model again? After linking the url "Hi, This is pretraind model(checkpoint)" you supported, I will got the following information 此项可能已被删除或过期,或者你没有查看权限。有关详细信息,请与此项的所有者联系。

SinDongHwan commented 3 years ago

hi, @Ai-is-light. I can't access my server pc. I will re-upload a pretrained model on next Monday(19th).

if somebody has a pretrained model, please share it.

tchennwpu commented 3 years ago

hi, @Ai-is-light. I can't access my server pc. I will re-upload a pretrained model on next Monday(19th).

if somebody has a pretrained model, please share it.

hi, could you please reshare the pre-trained model again?