Some problem when train Fusion

[nazimii]

Hello, I encountered some problems when retraining the fusion model. Some key parameter guidelines for training fusion are not given in the code warehouse. Can you provide it? Specifically as follows: (1) generate_fusion.py: parameter "separation" not given

Can you provide the relevant parameter descriptions of fusion training and the instructions to run so that I can reproduce the results of your paper?

and when I try to train（python train.py）,I meet some code mistake in fusion_dataset.py： （1）are there some mistake When you assign a value to self.vid_to_instance？ and It will return error at： self.videos = [v for v in self.videos if v in self.vid_to_instance]（line 60 in fusion_datast.py）

[nazimii]

@hkchengrex

[hkchengrex]

We provided the pre-computed fusion data for download. The names of the folders should reflect the parameters I used. Like sep30 means separation=30, m20 means mem_freq=20. I also haven't generated the fusion data for the entire dataset.

[hkchengrex]

My current L60 in fusion_dataset is different. I did a brief scan and couldn't find the problem.

[nazimii]

the problem when train fusion(train.py),is :

Traceback (most recent call last): File "train.py", line 103, in start_epoch = total_iter//len(train_loader) ZeroDivisionError: integer division or modulo by zero

so there must be something wrong at:

self.videos = [v for v in self.videos if v in self.vid_to_instance]（line 52 in fusion_datast.py）

and line 42~48 when you assign value to self.vid_to_instance

[hkchengrex]

So your self.vid_to_instance is empty? It doesn't happen to me.

[nazimii]

Really? I have to change the code (fusion_dataset.py, line 50) to : (fusion_dataset.py, line 50) for folder in fuse_list:

folder_path = path.join(self.fd_root, folder)

# video_list = sorted(os.listdir(folder_path))    # ['00000', '00020', '00040', '00060']
# print("video list", video_list)
# video level - different videos
# for vid in video_list:
video_path = path.join(self.fd_root, folder) #, vid)
self.vid_to_instance[folder].append(video_path)
total_fuse_vid += 1

so can I train the fusion model.
meybe the code at github is different from your's ?

[nazimii]

In my code the data in self.video is different from self.vid_to_instance.

so it comes error at fusion_dataset.py : self.videos = [v for v in self.videos if v in self.vid_to_instance]

[nazimii]

are you use YV when train the fusion model? how many dataset you used to trian fusion model?

[hkchengrex]

It seems like your folder structure is different, i.e., does not have the "run"-level. I used the uploaded "fusion_data" (a subset of BL30K + DAVIS) for training.

[nazimii]

hi, I have a new question:

can you give me the code that used to eval YTB dataset? (multi reference)

[hkchengrex]

https://github.com/hkchengrex/Mask-Propagation

[nazimii]

so, if I use the fusion_data you provide，I can get the result the same as paper?