Research map / Architectures and efficiency / CNN design
NetVLAD: CNN Architecture for Weakly Supervised Place Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train AlexNet NetVLAD end-to-end for place recognition rather than use off-the-shelf AlexNet VLAD | AlexNet VLAD representation · Pitts250k-test | recall@1: 55.0% -> 81.0% | Helped |
| Use NetVLAD instead of max pooling (ablation) | CNN place-recognition descriptor · Tokyo 24/7 | recall@1: 38.4% -> 42.9% | Helped |
| Train with Time Machine imagery rather than train without it (ablation) | AlexNet max-pooling place-recognition model · Pitts30k-val | recall@1: 38.7 -> 68.5 | Helped |
| Train the NetVLAD layer while keeping the CNN feature layers fixed (ablation) | AlexNet NetVLAD · Pitts30k-val | recall@1: 54.5 -> 80.5 | Helped |
| Train CNN layers down through conv3 rather than only the NetVLAD layer (ablation) | AlexNet NetVLAD · Pitts30k-val | recall@1: 80.5 -> 85.5 | Helped |
| Train layers below conv3 as well, through conv2 (ablation) | AlexNet NetVLAD · Pitts30k-val | recall@1: 85.5 -> 84.5 | Hurt |
| Train the full CNN through conv1 instead of stopping at conv3 (ablation) | AlexNet NetVLAD · Pitts30k-val | recall@1: 85.5 -> 84.2 | Hurt |
| Use Max pooling rather than NetVLAD pooling (ablation) | AlexNet place-recognition representation · Pitts30k-val | recall@1: 80.5 -> 68.5 | Hurt |
| Train with Time Machine data instead of using ImageNet-pretrained features (ablation) | AlexNet max-pooling model · Pitts30k-val | recall@1: 33.5 -> 68.5 | Helped |
| Train the Pittsburgh descriptor on urban data for retrieval benchmarks | VGG-16 NetVLAD · Holidays orig | mAP: 82.1 -> 79.9 | Hurt |
| Train VGG-16 NetVLAD on Pittsburgh instead of using its off-the-shelf representation | VGG-16 NetVLAD, 256-D · Oxford 5k crop | mAP: 55.5 -> 63.5 | Helped |
Closest papers by meaning
- BoQ: A Place is Worth a Bag of Learnable Queries (CVPR 2024 · similarity 0.88)
- SuperVLAD: Compact and Robust Image Descriptors for Visual Place Recognition (NeurIPS 2024 · similarity 0.88)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 2025 · similarity 0.87)
- Optimal Transport Aggregation for Visual Place Recognition (CVPR 2024 · similarity 0.87)
- Towards Implicit Aggregation: Robust Image Representation for Place Recognition in the Transformer Era (NeurIPS 2025 · similarity 0.87)
- CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective (CVPR 2024 · 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.