anthony-chaudhary
commented
3 years ago Hey @michaeljunkin Diffgram is a good alternative for labeling here! You can get started online and run it on your kubernetes cluster for max data control. https://diffgram.com/
Using the stand alone VOTT installed from the "exe" installer located here: https://github.com/Microsoft/VoTT/releases
I created two connections. One to a local file directory that contains 100 images. One to an empty directory. I've used the image directory as the source connection and the empty directory as the target directory.
I've also tried making the target and source the same image directory.
Either way after using those connections with a new project, the project opens, momentarily displays the images on a left hand pane, then displays a blank dark gray window. After that I have to close the window from the task bar - there are no controls on the blank window.
If I reload the project, the same thing happens.
The images are simple charts between 5 and 7 KB. Some are PNG some are JPG.
If I remove all the PNG files then it works and shows the JPG files.
The new object detection feature of the Microsoft Visual studio model builder seems to require a VOTT formatted JSON input file. Seem to suggest because they don't say that but they have a link to VOTT and recommend it as the tool to tag training data.
It's a shame they are linking to a tool that can't even open a PNG file. It's a shame that this is documented anywhere.