Will it work with DeepLabV3? about image-segmenter-ios HOT 7 OPEN

It should be ok, I have not tested. ues coremltools https://github.com/apple/coremltools

from image-segmenter-ios.

I have the Model (from Apple website), but the file UIImage+segmentation.swift needs ti ne readapted to use that model, some things have changed.

https://docs-assets.developer.apple.com/coreml/models/Image/ImageSegmentation/DeepLabV3/DeepLabV3.mlmodel

from image-segmenter-ios.

I adapted the code to use custom deeplabv3 converted with tfcoreml. You can see https://github.com/ismaeIfm/image-segmenter-ios

from image-segmenter-ios.

@ismaelfm is ths v3 model fast enough to be used for real time purposes?

from image-segmenter-ios.

@mevinDhun I don't think so, I've tested it on an iPhone XS and it takes a couple of seconds to perform the segmentation on the test image, but I'm not sure if some settings need to be changed to use the neural engine. I think a proper benchmarking is needed to answer properly.

from image-segmenter-ios.

I adapted the code to use custom deeplabv3 converted with tfcoreml. You can see https://github.com/ismaeIfm/image-segmenter-ios

Why don't you use the one apple gives you instead of converting it?

https://docs-assets.developer.apple.com/coreml/models/Image/ImageSegmentation/DeepLabV3/DeepLabV3.mlmodel

from image-segmenter-ios.

Because the Apple coreml model is “compiled” from input to predictions and this code assumes the output of the model is an internal layer (ResizeBilinear), so in order to make it work the code would need some changes and I'm not a swift developer.

from image-segmenter-ios.