search

csvt32745 / FTP-VM

GNU General Public License v3.0
21 stars 2 forks source link

FileNotFoundError: [Errno 2] No such file or directory: './data/VideoMatting108/frame_corr.json' #5

Open Aricling opened 1 year ago

Aricling commented 1 year ago

I did all the steps said in the tutorial. But when running model inference(inference_dataset.py) I got the error illustrated above. Has anyone also encountered the same issue? I don't know where the frame_corr.json is.

csvt32745 commented 1 year ago

Hi~ For VideoMatting108 dataset, you have to download frame_corr.json and other metadata in their drive (the same place as FG/BG). frame_corr.json provides the FG / BG pairs.

The expected file structure is

Aricling commented 1 year ago

Hi~ For VideoMatting108 dataset, you have to download frame_corr.json and other metadata in their drive (the same place as FG/BG). frame_corr.json provides the FG / BG pairs.

The expected file structure is

  • VideoMatting108

    • BG_done

    • (Video 0)

    • (Video 1)

    • ...

    • FG_done

    • ...

    • train_videos.txt

    • val_videos.txt

    • frame_corr.json

Okay, thank you, I solved it:) What's more, I saw that in your paper "Youtube-VIS [52], is adopted to train the trimap segmentation". But I didn't find the code generating trimap segmentation in this repo. Could you please offer me some hints that I can adopt to generate trimaps from a video? Thank you! :)

csvt32745 commented 1 year ago

The trimap ground truth is generated at https://github.com/csvt32745/FTP-VM/blob/main/dataset/youtubevis.py#L103 The unknown (gray) region comes from (the dilation - erosion) of masks. Since the segmentation mask is imperfect for trimap, the training on YT-VIS stops a bit earlier.