Feature extraction for a single image

[dillon-quan]

Hello,

I noticed that the package only works when extracting features for a directory of images but not for a single image. The error message that I get is shown down below. I was wondering if anyone can provide any suggestions on how to workaround this issue if I want to be able to generate feature extraction for a single image file. Thank you in advance!

...Creating dataset.
Traceback (most recent call last):
  File "/Users/dillonquan/Documents/heretic/feature_extraction.py", line 32, in <module>
    dataset = ImageDataset(
  File "<string>", line 9, in __init__
  File "/Users/dillonquan/anaconda3/envs/price/lib/python3.9/site-packages/thingsvision/utils/data/dataset.py", line 56, in __post_init__
    self._find_classes()
  File "/Users/dillonquan/anaconda3/envs/price/lib/python3.9/site-packages/thingsvision/utils/data/dataset.py", line 83, in _find_classes
    children = sorted([d.name for d in os.scandir(self.root) if d.is_dir()])
NotADirectoryError: [Errno 20] Not a directory: '/Users/dillonquan/Desktop/image_01.png'

[dillon-quan]

Closing the issue since I was able to make edits to the source code to allow for feature extraction on a single image.

[dillon-quan]

Reopening the issue here as a suggestion, I made edits to the source code on my local machine to allow feature extraction on a single image. The use case for this would be for performing image feature extraction in real time for model inference down stream. I am happy to contribute and add this to the package if I can get guidance on how to start or if thats something worth adding.

[LukasMut]

Reopening the issue here as a suggestion, I made edits to the source code on my local machine to allow feature extraction on a single image. The use case for this would be for performing image feature extraction in real time for model inference down stream. I am happy to contribute and add this to the package if I can get guidance on how to start or if thats something worth adding.

Hi Dilon! Thanks for your suggestion. That may indeed be a useful addition. Feel free to open a PR and see whether all tests are passing. Ideally, you'd add a unittest for that addition to the thingsvision tests. Please see the README and tests for how to do a PR and modify/add tests. I guess, the easiest would be to fork the repo. Cheers Lukas

[LukasMut]

@dillon-quan will you add a PR anytime soon?

[dillon-quan]

@LukasMut Thanks for the reminder. This has slipped my mind as my schedule has been busy. I dont think it will come anytime soon due to bandwidth. Im still hopeful that I can add this feature in.

[LukasMut]

@dillon-quan would you be able to finally submit that PR?

[Tankiit]

I did a dirty hack i.e squeezed to make sure it's a 4D tensor and it works. So anyone interested could potentially try this too.