search

autowarefoundation / autoware_ai

Apache License 2.0
20 stars 7 forks source link

Semantic segmentation using TensorFlow models #654

Closed drwnz closed 4 years ago

drwnz commented 5 years ago

Feature request

Description

Many semantic segmentation models are implemented in TensorFlow (Deeplab, ICNet, and so on). It would be useful to have a node for loading semantic segmentation models for inference on camera image topics.

Implementation considerations

Implemented with the TensorFlow wrapper from issue autowarefoundation/autoware_ai#653.

Alternatives

Semantic segmentation using ENet (implemented using caffe) is already implemented in computing/perception/detection/vision_detector/packages/vision_segment_enet_detect. The current state-of-the-art (DeepLabv3) is available pre-trained on Cityscapes in TensorFlow.

wew84 commented 5 years ago

I built a package that you might find interesting that opens freeze graphs of segmentation / detection and publishes them to ROS. https://github.com/wew84/cnn_bridge