Research map / Detection and segmentation / CNN features
Superpixel Convolutional Networks Using Bilateral Inceptions
Raghudeep Gadde, Varun Jampani, Martin Kiefel, Daniel Kappler, Peter V. Gehler
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add BI(6,2) module, training only the module | DeepLab · Pascal VOC12 test | IoU: 68.9 -> 70.8 | Improved |
| also train FC layers jointly with BI(6,2) (ablation) | DeepLab + BI(6,2) · Pascal VOC12 test | IoU: 70.8 -> 71.5 | Improved |
| use more kernels in the BI module (2 to 6 kernels at fc6) (ablation) | DeepLab + BI · Pascal VOC12 test | IoU: 71.5 -> 72.9 | Improved |
| stack two BI modules (6,2 and 7,6) instead of one (ablation) | DeepLab + BI · Pascal VOC12 test | IoU: 71.5 -> 73.6 | Improved |
| use 10 kernels at fc8 (BI(8,10)) instead of 6 kernels at fc7 (ablation) | DeepLab + BI · Pascal VOC12 test | IoU: 73.1 -> 72.0 | Hurt |
| finetune the entire network (FULL) rather than BI+FC only (ablation) | DeepLab + BI(6,2)-BI(7,6) · Pascal VOC12 test | IoU: 73.6 -> 74.1 | Improved |
| add DenseCRF on top of the full BI model (ablation) | DeepLab + BI(6,2)-BI(7,6) · Pascal VOC12 test | IoU: 74.1 -> 75.1 | Improved |
| replace DenseCRF post-processing with BI(6,2)-BI(7,6) modules | DeepLab · Pascal VOC12 test | IoU: 72.7 -> 73.6 | Improved |
| replace deconvolution and DenseCRF with BI modules | DeconvNet / CRFasRNN · Pascal VOC12 test | IoU: 74.7 -> 74.9 | Improved |
| insert BI modules after fc7 and fc8 of AlexNet | AlexNet · MINC test | class accuracy: 55.3 -> 69.4 | Improved |
| replace DenseCRF with BI modules | AlexNet · MINC test | total accuracy: 71.0 -> 72.8 | Improved |
| insert BI(6,2)-BI(7,6) modules | DeepLab · Cityscapes val (full-res.) | IoU: 65.7 -> 66.9 | Improved |
Closest papers by meaning
- Superpixel Segmentation With Fully Convolutional Networks (CVPR 2020 · similarity 0.88)
- STD2P: RGBD Semantic Segmentation Using Spatio-Temporal Data-Driven Pooling (CVPR 2017 · similarity 0.87)
- Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells (CVPR 2019 · similarity 0.86)
- Weakly Supervised Semantic Segmentation Using Superpixel Pooling Network (AAAI 2017 · similarity 0.86)
- Segmentation-Aware Convolutional Networks Using Local Attention Masks (ICCV 2017 · similarity 0.86)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
