Would it be possible to integrate object detection or image segmentation models?

[swiss-knight]

Hello,

I'm wondering in a fully open minded perspective if there is some place for keras models (or at least, their architecture) related to these "family of methods" in the "object detection and/or image segmentation" space: [ 'Fully Convolutional', 'Region-based', 'Single shot' ] such as:

Original Fully ConvNet paper:

• FCN-8s, FCN-16s, FCN-32s from https://arxiv.org/abs/1411.4038

Region-based methods:

• R-FCN from https://arxiv.org/pdf/1605.06409
• R-CNN from https://arxiv.org/pdf/1311.2524
• Fast R-CNN from https://arxiv.org/abs/1504.08083
• Faster R-CNN from https://arxiv.org/pdf/1506.01497
• Mask R-CNN from https://arxiv.org/abs/1703.06870
• SPP-net from https://arxiv.org/abs/1406.4729

And the, let's say, "one shot" family:

• SSD from https://arxiv.org/abs/1512.02325
• YOLO from https://arxiv.org/abs/1506.02640
• YOLO v2 (a.k.a. know as YOLO9000) from https://arxiv.org/abs/1612.08242
• YOLO v3 from https://arxiv.org/abs/1804.02767

Probably others derived methods, but let's stick on these well known papers for the moment.

Thanks.

· If some work to try implementing them has already been taken or done / is actually taking place, or if it's purely and technically not possible for some reason (for the moment), I would be glad to know.
· Also feel free to redirect me to more pertinent place if this one is not (I'm still in a 'discovering deep learning' stage).