Research map / Label-efficient and robust learning / Recognition and long-tail
Recurrent Image Annotation With Explicit Inter-Label Dependencies
Ayushi Dutta, Yashaswi Verma, C.V. Jawahar
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 |
|---|---|---|---|
| replace frequent-first label ordering with multi-order training (mask previously chosen label) | CNN-RNN · MS-COCO val | per-label F1: 62.63 -> 70.13 | Improved |
| replace frequent-first label ordering with multi-order training | CNN-RNN · MS-COCO val | per-image F1: 68.13 -> 80.11 | Improved |
| replace rare-first ordering with multi-order training | CNN-RNN · MS-COCO val | per-label F1: 63.43 -> 70.13 | Improved |
| add an RNN on top of the CNN (frequent-first order) | CNN · MS-COCO val | per-image F1: 63.30 -> 68.13 | Improved |
| use rare-first instead of frequent-first label order | CNN-RNN · MS-COCO val | per-image F1: 68.13 -> 67.64 | Hurt |
| use rare-first instead of frequent-first label order | CNN-RNN · MS-COCO val | per-label F1: 62.63 -> 63.43 | Improved |
| replace candidate-label-pool order-free training with multi-order training | Order-free RNN · MS-COCO val | per-image F1: 67.70 -> 80.11 | Improved |
| replace rare-to-frequent training with multi-order training | SR-CNN-RNN · MS-COCO val | per-image F1: 72.47 -> 80.11 | Improved |
| replace rare-to-frequent training with multi-order training | SR-CNN-RNN · MS-COCO val | top-1 accuracy: 81.44 -> 93.49 | Improved |
| replace rare-to-frequent training with multi-order training | SR-CNN-RNN · NUS-WIDE test | top-1 accuracy: 68.06 -> 84.05 | Improved |
| replace PLA with multi-order training | PLA · NUS-WIDE test | per-image F1: 72.37 -> 74.74 | Improved |
| compare Multi-order RNN with ACfs | Multi-order RNN · MS-COCO val | per-label F1: 72.20 -> 70.13 | Hurt |
Closest papers by meaning
- Semantic Regularisation for Recurrent Image Annotation (CVPR 2017 · similarity 0.92)
- CNN-RNN: A Unified Framework for Multi-Label Image Classification (CVPR 2016 · similarity 0.91)
- Orderless Recurrent Models for Multi-Label Classification (CVPR 2020 · similarity 0.88)
- Order-Free RNN With Visual Attention for Multi-Label Classification (AAAI 2018 · similarity 0.88)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.87)
- Pre-Trained Vision-Language Models as Noisy Partial Annotators (AAAI 2025 · 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.
