search

alexander-pv / maskrcnn_tf2

Mask R-CNN for object detection and instance segmentation with Keras and TensorFlow V2 and ONNX and TensorRT optimization support.
Other
40 stars 11 forks source link

fpn mask layers #4

Open VeeranjaneyuluToka opened 2 years ago

VeeranjaneyuluToka commented 2 years ago

Hi, I felt you have created a very modularized implementation and it is easy to understand. I have one question for fpn mask layers which is fpn_mask_graph() in your implementation. I think you have used same layers irrespective of backbone type, is my understanding correct. If yes, do you think that is fine. Do not we need to add the respective backbone blocks in fpn mask also?

Thanks, Veeru.

alexander-pv commented 2 years ago

Hi, @VeeranjaneyuluToka , I'm glad that you found this repository useful!

You are right. Generally, I refactored the original matterport model realization and started to add small improvements to it. So, now for all backbones /src/config.py holds the same params.

I believe that each backbone can be exploited more efficiently with new c2-c5 features. Perhaps, some simple tasks can be solved even with a smaller group of features.

I believe this is definitely important research that concerns cost complexity as much as model accuracy. I added "Flexible backbones configuration" point in TODOs section of README.md. If you are interested in such improvements, you can send PRs for collaboration. Looking at the model basic scheme It seems to me that flexible backbones configuration will also require code structure improvements.

mask_rcnn_training_mode