Segmentation models?

[InternetMaster1]

Wow what an amazing amazing library!!!

Do you have pre-trained models for image segmentation?

I am a newbie. These models seem to be for image classification, correct?

[pfeatherstone]

https://github.com/qubvel/segmentation_models.pytorch

[rwightman]

@InternetMaster1 thanks! ... I was going to recommend what @pfeatherstone already posted, qubvel's collection is really good for segmentation models in PyTorch. The models here are currently all classification. I've actually used qubvel's unet and deeplab arch with the pretrained backbone models here and an evolved version of my training code on some non open-source projects.

I hope to find time someday to bring some obj detection or segmentation into this repo but that won't be any time soon. I've been working on EfficientDet recently, and intend to try leveraging that for semantic segmetnation, it uses this code as a dependency. https://github.com/rwightman/efficientdet-pytorch

[InternetMaster1]

@pfeatherstone Many thanks!

@rwightman

Many thanks for the detailed reply! Good luck with EfficientDet library. It looks to be making amazing progress. I am new to segmentation, and had a couple of questions, could you help me out?

I am looking to implement high-quality semantic segmentation on a mobile device for human segmentation.

Would EfficientDet be a good option for the same? How does it compare to MobileNetV3 or maybe SINet/ExtremeC3Net?

For training of human segmentation, I was planning to use the Supervisely Person dataset.

In the final output mask, how can I even get the objects that a person is holding, say a cup, a purse, a tennis racquet, a toy, a magazine. It could be just about anything.

I am very much perplexed with this problem.

If I am not mistaken, the Supervisely dataset doesn't contain masks for objects that the person might be holding. To achieve this, would a dataset like Supervisely be unfit for the job? Or we need to train on a dataset with more labels than just "person"?

But ideally, if an object is lying on the side, it is ok if it does not come in the mask. But if the person is holding the object, it should definitely come in the final mask.

How can this be achieved?