High-level Feature Guided Decoding for Semantic Segmentation
This work addresses accuracy issues in semantic segmentation for computer vision applications, offering an incremental improvement over existing methods.
The paper tackles the problem of contaminated high-level features in pyramid-based upsamplers for semantic segmentation, which leads to less accurate results compared to dilation-based models, and proposes a High-level Feature Guided Decoder (HFGD) that uses pre-trained high-level features as guidance to achieve state-of-the-art results on benchmarks like Pascal Context, COCOStuff164k, and Cityscapes without extra training data.
Existing pyramid-based upsamplers (e.g. SemanticFPN), although efficient, usually produce less accurate results compared to dilation-based models when using the same backbone. This is partially caused by the contaminated high-level features since they are fused and fine-tuned with noisy low-level features on limited data. To address this issue, we propose to use powerful pre-trained high-level features as guidance (HFG) so that the upsampler can produce robust results. Specifically, \emph{only} the high-level features from the backbone are used to train the class tokens, which are then reused by the upsampler for classification, guiding the upsampler features to more discriminative backbone features. One crucial design of the HFG is to protect the high-level features from being contaminated by using proper stop-gradient operations so that the backbone does not update according to the noisy gradient from the upsampler. To push the upper limit of HFG, we introduce a context augmentation encoder (CAE) that can efficiently and effectively operate on the low-resolution high-level feature, resulting in improved representation and thus better guidance. We named our complete solution as the High-Level Features Guided Decoder (HFGD). We evaluate the proposed HFGD on three benchmarks: Pascal Context, COCOStuff164k, and Cityscapes. HFGD achieves state-of-the-art results among methods that do not use extra training data, demonstrating its effectiveness and generalization ability.