Research map / Architectures and efficiency / Attention and transformers
ParC-Net: Position Aware Circular Convolution with Merits from ConvNets and Transformer
Haokui Zhang, Wenze Hu, Xiaoyu Wang
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 position embedding to ParC (ablation) | ParC-Net-S · MS-COCO detection | mAP: 27.5 -> 28.5 | Improved |
| add position embedding to ParC (ablation) | ParC-Net-S · PASCAL VOC segmentation | mIOU: 79.2 -> 79.7 | Improved |
| add position embedding to ParC (ablation) | ParC-Net-S · ImageNet-1k classification | top-1: 78.63 -> 78.63 | No clear effect |
| use meta-former architecture instead of ResNeXt block (ablation) | ParC-Net-S · ImageNet-1k classification | top-1: 76.00 -> 78.63 | Improved |
| add channel-wise attention (ablation) | ParC-Net-S · ImageNet-1k classification | top-1: 78.50 -> 78.63 | Improved |
| use ParC instead of big kernel 1/4 (ablation) | ParC-Net-S · ImageNet-1k classification | top-1: 78.46 -> 78.63 | Improved |
| increase big kernel size from 1/4 to 1/2 of input (ablation) | big-kernel variant · ImageNet-1k classification | top-1: 78.46 -> 78.45 | No clear effect |
| replace ViT blocks of MobileViT-S with ParC blocks | MobileViT-S · ImageNet-1k classification | top-1 / params: 78.4 -> 78.6; 5.6M -> 5.0M params | Improved |
| replace ViT blocks with ParC blocks (RK3288 latency) | MobileViT-S · RK3288 latency | ms: 457 -> 353 | Improved |
| replace ViT blocks with ParC blocks (DP2000 latency) | MobileViT-S · DP2000 latency | ms: 368 -> 98 | Improved |
| insert ParC into ResNet50 | ResNet50 · ImageNet-1k classification | top-1: 78.8 -> 79.6 | Improved |
| insert ParC into MobileNetV2 | MobileNetV2 · ImageNet-1k classification | top-1: 70.2 -> 71.1; latency 24 -> 27 ms | Mixed |
Closest papers by meaning
- ParCNetV2: Oversized Kernel with Enhanced Attention (ICCV 2023 · similarity 0.89)
- CvT: Introducing Convolutions to Vision Transformers (ICCV 2021 · similarity 0.88)
- Conv2Former: A Simple Transformer-Style ConvNet for Visual Recognition (TPAMI 2024 · similarity 0.88)
- ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases (ICML 2021 · similarity 0.88)
- CMT: Convolutional Neural Networks Meet Vision Transformers (CVPR 2022 · similarity 0.87)
- ConvNet vs Transformer, Supervised vs CLIP: Beyond ImageNet Accuracy (ICML 2024 · similarity 0.87)
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.
