Import third party models

Hello, I was reading the new docs of Deepstack and I couldn’t find the documentation to import tensorflow models anymore, so I’m thinking that maybe that option is not available in the new version?

It’s very nice that now it’s easier to train with custom images, but the problem for me is that labeling images one by one it’s very time consuming and even impossible.
I wanted to use the MaskedFace-Net database to create a model that can detect if a mask is being used correctly or not. But it has 137,000 images so it looks like impossible to label that many images.
With tensorflow you can just use the make_image_classifier tool where there is no need to manually label images, you can just create a folder for each label and that’s it. That is the only reason I’m thinking about using TF.

Deepstack have been working very well for me. I’m using windows 10 so I’m eager for the new update.


Hello @mrpie. Thanks for bringing this up. We did removed the custom recognition feature with the aim of bringing back an easier approach to do this.

Before Christmas, we shall release the ability to put your images in a folder and just train it with DeepStack trainer and deploy your image recognition model easily
Stay tuned, this is in progress and will be announced soon.


Hi John. I bet you’ve been very busy but I would like to know, is this feature still on your list?
I’ve gathered big databases of thousands of pictures for different ideas, so that means I can’t label each image.

When I was doing that I saw your fire-flame-dataset, it was exactly what I was looking for. But I’m waiting for this feature in order to train a new model, as manually labeling would be very slow.

1 Like

Hello @mrpie, this feature is still on our list and would likely be released this month. Thanks for your patience.

1 Like

Any news on this topic? I need to import this frozen model: MegaDetector