Research map / Detection and segmentation / CNN features
SaccadeNet: A Fast and Accurate Object Detector
Shiyi Lan, Zhou Ren, Yi Wu, Larry S. Davis, Gang Hua
Check 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 corner keypoint supervision through Corner-Attn (ablation) | SaccadeNet-ResNet-18 · PASCAL VOC | mAP@50: 70.69 -> 71.02 (+0.33) | Improved |
| add corner keypoint supervision through Corner-Attn (ablation) | SaccadeNet-ResNet-18 · PASCAL VOC | mAP@90: 16.48 -> 13.51 (-2.97) | Hurt |
| add Aggregation-Attn for boundary refinement (ablation) | SaccadeNet · MS COCO val2017 | mAP: 37.9 -> 38.8 | Improved |
| add both Corner-Attn and Aggregation-Attn (ablation) | SaccadeNet-ResNet-18 · PASCAL VOC | mAP@90: 16.48 -> 21.07 (+4.59) | Improved |
| replace corners with diagonal points at 0.8 (ablation) | SaccadeNet-ResNet-18 keypoint inputs · PASCAL VOC | mAP@90: 21.07 -> 18.27 | Hurt |
| replace corners with middle-edge points (ablation) | SaccadeNet-ResNet-18 keypoint inputs · PASCAL VOC | mAP@70: 57.84 -> 55.85 | Hurt |
| increase boundary refinement from one to two iterations (ablation) | SaccadeNet-ResNet-18 · PASCAL VOC | mAP@90: 21.07 -> 21.32 | Improved |
| add Aggregation-Attn-Cls to refine classification scores (ablation) | SaccadeNet-ResNet-18 · PASCAL VOC | mAP@50: 70.92 -> 52.26 | Hurt |
| use center features alone instead of combining center and corner features (ablation) | SaccadeNet Aggregation-Attn · PASCAL VOC | mAP@90: 21.07 -> 19.01 | Hurt |
| use corner features alone instead of combining center and corner features (ablation) | SaccadeNet Aggregation-Attn · PASCAL VOC | mAP@90: 19.78 -> 21.07 | Hurt |
| use flip testing (ablation) | SaccadeNet-DLA · MS COCO val2017 | mAP: 38.8 -> 40.7 | Improved |
| replace peak-picking NMS with IoU-based NMS (ablation) | SaccadeNet-DLA · MS COCO val2017 | mAP: 38.8 -> 39.3 | Improved |
Closest papers by meaning
- Saccader: Improving Accuracy of Hard Attention Models for Vision (NeurIPS 2019 · similarity 0.87)
- Improved Efficiency Based on Learned Saccade and Continuous Scene Reconstruction From Foveated Visual Sampling (ICLR 2024 · similarity 0.87)
- HoughNet: Integrating Near and Long-Range Evidence for Visual Detection (TPAMI 2022 · similarity 0.86)
- Learning Rich Features at High-Speed for Single-Shot Object Detection (ICCV 2019 · similarity 0.86)
- Object-aware Gaze Target Detection (ICCV 2023 · similarity 0.86)
- ObjectNet: A large-scale bias-controlled dataset for pushing the limits of object recognition models (NeurIPS 2019 · similarity 0.86)
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.
