Feature Fusion/Integration

1) Dataset:

2) Papers:

Paper List:

1. AttaNet: Attention-Augmented Network for Fast and Accurate Scene Parsing [Paper] [Personal Summary]

2. Deep Layer Aggregation_CVPR_2018 [Paper]

3. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation_ECCV_2018 [Paper]

4. 
   

Reading List:

• [Tony] Small Object Detection using Context and Attention_arXiv_2019 [Link]

• [Tony] MultiResolution Attention Extractor for Small Object Detection_arXiv_2020 [Link]

• 
  

To-do List:

Combine multi-level representations:

1. • Fully convolutional networks for semantic segmentation_CVPR_2015

   • Convolutional networks for biomedical image segmentation_ICM_2005

   • Refinenet: Multi-path refinement networks for high-resolution semantic segmentation_CVPR_2017

   • Segnet: A deep convolutional encoder-decoder architecture for image segmentation_PAMI_2017

   • Adaptive context network for scene parsing_ICCV_2019

Adopt the multi-branch framework (an extra branch)

1. • [ICNet]_Icnet for real-time semantic segmentation on high-resolution images_ECCV_2018

   • Bisenet: Bilateral segmentation network for realtime semantic segmentation_ECCV_2018

   • BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation_2020

Implement a cross-level feature aggregation architecture with less computation. 

1. • Dfanet: Deep feature aggregation for real-time semantic segmentation_CVPR_2019

   • Semantic Flow for Fast and Accurate Scene Parsing_arXiv_2020

Uses gates to control information propagation

1. • GFF: Gated Fully Fusion for Semantic Segmentation_arXiv_2019.

Adopts a lightweight attention strategy 

1. • AttaNet: Attention-Augmented Network for Fast and Accurate Scene Parsing_AAAI_2020

Strengthen feature integration: 

• Skip connection [51], hyper-column [22]: Integrate low-level physical feature to high-level semantic feature. 

• SFAM [98], ASFF [48], and BiFPN [77]: Integrate different feature pyramid (ideas from FPN)

1. SFAM [98]: Use SE module to execute channel-wise level re-weighting on multi-scale concatenated feature maps. 

2. ASFF [48]: Use soft-max as point-wise level reweighting and then adds feature maps of different scales.

3. BiFPN [77]: Use multi-input weighted residual connections to execute scale-wise level re-weighting, and then add feature maps of different scales

Unfilter:

https://arxiv.org/pdf/2103.10643.pdf

3) References: