Research map / Detection and segmentation / Open-set and few-shot detection
Hierarchy Parsing for Image Captioning
Ting Yao, Yingwei Pan, Yehao Li, Tao Mei
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 hierarchy parsing (HIP) features | Up-Down · COCO Karpathy test (cross-entropy) | CIDEr-D: 113.5 -> 116.6 | Improved |
| add hierarchy parsing (HIP) features | GCN-LSTM · COCO Karpathy test (cross-entropy) | CIDEr-D: 117.1 -> 120.3 | Improved |
| add HIP features | Up-Down · COCO Karpathy test (CIDEr-D optimization) | CIDEr-D: 120.1 -> 127.2 | Improved |
| add HIP features | GCN-LSTM · COCO Karpathy test (CIDEr-D optimization) | CIDEr-D: 128.7 -> 130.6 | Improved |
| add Tree-LSTM-enhanced features to region features (ablation) | Up-Down · COCO Karpathy test (cross-entropy, Up-Down decoder) | CIDEr-D: 113.5 -> 115.5 | Improved |
| add instance-level features to region features (ablation) | Up-Down · COCO Karpathy test (cross-entropy, Up-Down decoder) | CIDEr-D: 113.5 -> 114.9 | Improved |
| add region features to instance and Tree-LSTM features (full HIP) (ablation) | Up-Down · COCO Karpathy test (cross-entropy, Up-Down decoder) | CIDEr-D: 115.2 -> 116.6 | Improved |
| use instance-level features alone instead of region features (ablation) | Up-Down · COCO Karpathy test (cross-entropy, Up-Down decoder) | CIDEr-D: 113.5 -> 113.3 | Hurt |
| use HIP image-level features for recognition | Up-Down features · COCO multi-label recognition | C-F1: 63.74 -> 65.74 | Improved |
| use HIP image-level features for recognition | Up-Down features · COCO multi-label recognition | O-F1: 65.41 -> 68.61 | Improved |
| add HIP to GCN-LSTM | GCN-LSTM · COCO online test server (c5) | CIDEr-D: 125.3 -> 127.9 | Improved |
| add HIP to GCN-LSTM | GCN-LSTM · Human evaluation (1K COCO test images) | M1: 73.9% -> 76.5% | Improved |
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.
- RFNet — RF-Net: An End-To-End Image Matching Network Based on Receptive Field (CVPR 2019) · better on 3 of 3 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
- Hierarchical Scene Parsing by Weakly Supervised Learning with Image Descriptions (TPAMI 2018 · similarity 0.88)
- Hierarchical Attention Network for Image Captioning (AAAI 2019 · similarity 0.88)
- Deep Structured Scene Parsing by Learning With Image Descriptions (CVPR 2016 · similarity 0.88)
- From Show to Tell: A Survey on Deep Learning-Based Image Captioning (TPAMI 2022 · similarity 0.87)
- Convolutional Image Captioning (CVPR 2018 · similarity 0.87)
- Dense Captioning With Joint Inference and Visual Context (CVPR 2017 · 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.
