Research map / Detection and segmentation / CNN features
ESPNet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation
Sachin Mehta, Mohammad Rastegari, Anat Caspi, Linda Shapiro, Hannaneh Hajishirzi
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 |
|---|---|---|---|
| use PReLU instead of ReLU (ablation) | ESPNet-A · Cityscapes val | mIoU: 0.36 -> 0.38 | Improved |
| add residual skip-connections in ESP modules (ablation) | ESPNet-A · Cityscapes val | mIoU: 0.37 -> 0.39 | Improved |
| use strided ESP instead of strided convolution for downsampling (ablation) | ESPNet-A · Cityscapes val | mIoU: 0.38 -> 0.39 | Improved |
| raise width divider K from 2 to 8 (ablation) | ESPNet-A · Cityscapes val | mIoU: 0.415 -> 0.303 | Hurt |
| concatenate feature maps for width expansion (ESPNet-B) (ablation) | ESPNet-A · Cityscapes val | mIoU: 0.39 -> 0.40 | Improved |
| add input reinforcement (ESPNet-C) (ablation) | ESPNet-B · Cityscapes val | mIoU: 0.40 -> 0.42 | Improved |
| replace input reinforcement with input-aware fusion (ablation) | ESPNet-C · Cityscapes val | mIoU: 0.42 -> 0.42 | No clear effect |
| add light-weight decoder to ESPNet-C (ablation) | ESPNet-C (alpha3=3) · Cityscapes val | mIoU (%): 49.0 -> 56.3 | Improved |
| use point-wise reduce instead of 3x3 standard convolution (ablation) | ESP module · Cityscapes val | mIoU (%): 50.8 -> 49.0 | Hurt |
| use dilated convolutions instead of standard 3x3 in spatial pyramid (ablation) | ESP module · Cityscapes val | mIoU (%): 47.4 -> 49.0 | Improved |
| increase depth multiplier alpha3 from 3 to 8 (ablation) | ESPNet · Cityscapes val | mIoU (%): 56.3 -> 61.4 | Improved |
| use ESP module instead of MobileNet module in ESPNet-C | ESPNet-C · Cityscapes val | mIoU: 7% higher | Improved |
Later papers that compared with this method
3 later accepted papers report ESPNetv2 as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Dual Super-Resolution Learning for Semantic Segmentation (CVPR 2020 · better on 3 of 3 shared results)
- DiCENet: Dimension-Wise Convolutions for Efficient Networks (TPAMI 2020 · better on 3 of 3 shared results)
- Highly Efficient Salient Object Detection with 100K Parameters (ECCV 2020 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SegNet — SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation (TPAMI 2017) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ESPNetv2: A Light-Weight, Power Efficient, and General Purpose Convolutional Neural Network (CVPR 2019 · similarity 0.90)
- DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation (CVPR 2019 · similarity 0.86)
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (TPAMI 2017 · similarity 0.86)
- FasterSeg: Searching for Faster Real-time Semantic Segmentation (ICLR 2020 · similarity 0.85)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 2024 · similarity 0.84)
- nnWNet: Rethinking the Use of Transformers in Biomedical Image Segmentation and Calling for a Unified Evaluation Benchmark (CVPR 2025 · similarity 0.83)
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.
