Own Dataset

[thomasdelteil]

Own Dataset

[ThomasDelteil]

Effectively you can remove all the download file logic etc, and then you put the prefix of your jpg images for example like this:

image_paths = '<HERE GOES YOUR PATH>'
list_files = glob.glob(os.path.join(images_path, '**.jpg'))

The only constraint it has is to have all your images in a single folder.

[ThomasDelteil]

this approach uses KNN search on the deep features of the test image for inference.
Look at the code here: https://github.com/ThomasDelteil/VisualSearch_MXNet/blob/master/mms/visualservice.py#L42

[visuazn]

Thanks for that.
how about Creating data files for offline inference?
do i need to work with this same?

[visuazn]

Cool Thanks. Will check this