IsADirectoryError when loading dataset

[p4bloR]

Hi, I'm currently trying to follow this tutorial: https://towardsdatascience.com/custom-instance-segmentation-training-with-7-lines-of-code-ff340851e99b
The problem comes when I try to load de dataset in this line of code:

vis_img.load_dataset("Nature")

I get the following error:

IsADirectoryError: [Errno 21] Is a directory: '/content/drive/MyDrive/Test_2/Nature/train.json'

I tried making my own dataset with labelme and the error persists. I'm currently running this on a google colab machine with a conda enviroment running.
I did some research and it seems this error comes from trying to pass a directory when python expects a file, but I don't understand what I should do differently, I'm literally copying code from a tutorial. Maybe I should try with older versions of the same libraries?
Thanks in advance.

[Dušan Radivojević]

I have same problem too.

[p4bloR]

Reading other threads it seemed that this problem is caused by an update in the labelme2coco library. So this problem can be fixed by using older libraries.

I copied this code from another issue and it solved it for me:

!pip3 uninstall tensorflow
!pip3 install tensorflow==2.4.1
!pip3 install tensorflow--gpu
!pip3 install imgaug
!pip install pixellib==0.5.2
!pip install labelme2coco==0.1.2

EDIT: This worked for me on google colab.

[Dušan Radivojević]

Thanks p4bloR. In my case it works:
pixellib==0.7.0
labelme2coco==0.1.2
but only on colab. For my Windows10 and on Kaggle also don't work.
Do somebody now how to do prediction with instance_custom_training? Some code example...

[p4bloR]

Thanks p4bloR. In my case it works: pixellib==0.7.0 labelme2coco==0.1.2 but only on colab. For my Windows10 and on Kaggle also don't work. Do somebody now how to do prediction with instance_custom_training? Some code example...

You're welcome, thanks for sharing what worked for you.

About the instance_custom_training thing, I think you should open a new issue to ask about it. This issue is about receiving a IsADirectoryError error when trying to load a dataset. I think it's important to keep issues like this organized so people in the future can quickly search and find answers.

[stephanietheanalyst]

I'm having the same problem. It's a csv and I have no problems in colab but it's giving me a directory error.

[juancarloscg]

Hi all, I am still incapable of making it work. I am using Colab with this environment:

!pip3 uninstall tensorflow !pip3 install tensorflow==2.4.3 !pip3 install tensorflow--gpu !pip3 install imgaug !pip install pixellib==0.5.2 !pip install labelme2coco==0.1.2

When I run this:

import pixellib from pixellib.custom_train import instance_custom_training train_maskrcnn = instance_custom_training() train_maskrcnn.modelConfig(network_backbone = "resnet101", num_classes= 1, batch_size = 2) train_maskrcnn.load_pretrained_model("/content/drive/My Drive/Colab Notebooks/mask_rcnn_coco.h5") train_maskrcnn.load_dataset("/content/drive/My Drive/Colab Notebooks/Cars") train_maskrcnn.train_model(num_epochs = 300, augmentation=True, path_trained_models = "mask_rcnn_models")

I get this

IsADirectoryError: [Errno 21] Is a directory: '/content/drive/My Drive/Colab Notebooks/Cars/train.json'

It seems that instead of creating both train and test jsons in their respective directory, it is creating just the training one and providing the path of a folder instead of the json object which is inside that folder and named dataset.json.

Could anyone help?

Thanks in advance.

[Hitomamacs]

Hi all, I am still incapable of making it work. I am using Colab with this environment:

!pip3 uninstall tensorflow !pip3 install tensorflow==2.4.3 !pip3 install tensorflow--gpu !pip3 install imgaug !pip install pixellib==0.5.2 !pip install labelme2coco==0.1.2

When I run this:

import pixellib from pixellib.custom_train import instance_custom_training train_maskrcnn = instance_custom_training() train_maskrcnn.modelConfig(network_backbone = "resnet101", num_classes= 1, batch_size = 2) train_maskrcnn.load_pretrained_model("/content/drive/My Drive/Colab Notebooks/mask_rcnn_coco.h5") train_maskrcnn.load_dataset("/content/drive/My Drive/Colab Notebooks/Cars") train_maskrcnn.train_model(num_epochs = 300, augmentation=True, path_trained_models = "mask_rcnn_models")

I get this

IsADirectoryError: [Errno 21] Is a directory: '/content/drive/My Drive/Colab Notebooks/Cars/train.json'

It seems that instead of creating both train and test jsons in their respective directory, it is creating just the training one and providing the path of a folder instead of the json object which is inside that folder and named dataset.json.

Could anyone help?

Thanks in advance.

Same here does somebody has a solution?

[Zhang Jian]

!pip3 uninstall tensorflow
!pip3 install tensorflow==2.4.1
!pip3 install tensorflow--gpu
!pip3 install imgaug
!pip install pixellib==0.5.2
!pip install labelme2coco==0.1.0

worked for me.

The directory error is caused by PixelLib/pixellib/custom_train/init.py /, line 46, labelme2coco.convert(labelme_folder1, save_json_path1).

Downgrading the labelme2coco library solves the directory error.

[ezcs software developer]

I am experiencing this issue myself too and, I would rather not downgrade the version of the libraries. Is this issue being troubleshooted? The issue seems to be pretty much understood. The new labelme2coco is generating a dataset.json file inside a subfolder that on previous version, that same folder name used to be the json file itself. Seems a few lines of code need to be added in PixelLib to set the labelme2coco destination path for the json file. Any date when this issue will be fixed? I will really appreciate! :)

[Emilie Zawadzki]

I agree with @ezcsdeveloper , I would rather not downgrade too.
I still have this issue in this configuration :

pixellib==0.7.1
labelme2coco==0.2.4
tensorflow==2.4.1 gpu_py39h8236f22_0
tensorflow-base==2.4.1 gpu_py39h29c2da4_0
tensorflow-estimator==2.6.0 pyh7b7c402_0
tensorflow-gpu==2.4.1 h30adc30_0

Python 3.9.16

Here is my Traceback :

Traceback (most recent call last):
  File "/home/pixellib-webcam/main2.py", line 13, in <module>
    vis_img.load_dataset("mydatasetfolder")
  File "/home/anaconda3/envs/tf-gpu/lib/python3.9/site-packages/pixellib/custom_train/__init__.py", line 68, in load_dataset
    self.dataset_train.load_data(save_json_path1, labelme_folder1)
  File "/home/anaconda3/envs/tf-gpu/lib/python3.9/site-packages/pixellib/custom_train/__init__.py", line 175, in load_data
    json_file = open(annotation_json)
IsADirectoryError: [Errno 21] Is a directory: '/home/pixellib-webcam/mydatasetfolder/train.json'

[Amadej Šenk]

Basically, you have to downgrade labelme2coco.

Run the following command:

!pip install pixellib==0.5.2
!pip install labelme2coco==0.1.0

After this, click to "Restart runtime", delete the directory train.json running rm-rf path-to-/train.json and try again.