Research map / Detection and segmentation / CNN features
Incorporating Copying Mechanism in Image Captioning for Learning Novel Objects
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 |
|---|---|---|---|
| Add copying from detected objects to standard generation | LSTM-C versus NOC · held-out MSCOCO | F1 average: 50.97 -> 54.40 | Helped |
| Add copying from detected objects to standard generation | LSTM-C versus NOC · held-out MSCOCO | F1 average: 51.85 -> 55.66 | Helped |
| Add copying from detected objects to standard generation | LSTM-C versus NOC · ImageNet, MSCOCO text | Accuracy: 10.04 -> 11.83 | Helped |
| Add copying from detected objects to standard generation | LSTM-C versus NOC · ImageNet, MSCOCO text | F1: 15.63 -> 16.39 | Helped |
| Add copying from detected objects to standard generation | LSTM-C versus NOC · ImageNet, BNC&Wiki text | Accuracy: 21.96 -> 31.11 | Helped |
| Use One hot+Glove word representations instead of One hot (ablation) | LSTM-C · held-out MSCOCO | F1 average: 54.40 -> 55.66 | Helped |
| Use One hot+Glove word representations instead of One hot (ablation) | LSTM-C · held-out MSCOCO | METEOR: 22 -> 23 | Helped |
| Use One hot+Glove word representations instead of One hot (ablation) | LSTM-C · held-out MSCOCO, couch | F1: 26.01 -> 38.77 | Mixed |
| Use One hot+Glove word representations instead of One hot (ablation) | LSTM-C · held-out MSCOCO, microwave | F1: 26.04 -> 27.81 | Helped |
| Set the generative-copying tradeoff parameter λ near 0.2 (ablation) | LSTM-C (One hot+Glove) · held-out MSCOCO | F1 | Helped |
| Increase λ beyond 0.6 (ablation) | LSTM-C (One hot+Glove) · held-out MSCOCO | F1 | Hurt |
| Add external BNC and Wikipedia text data (ablation) | LSTM-C (One hot+Glove) · ImageNet | Accuracy: 11.83 -> 31.11 | Helped |
Closest papers by meaning
- Pointing Novel Objects in Image Captioning (CVPR 2019 · similarity 0.93)
- Convolutional Image Captioning (CVPR 2018 · similarity 0.91)
- Reference Based LSTM for Image Captioning (AAAI 2017 · similarity 0.90)
- Deep Compositional Captioning: Describing Novel Object Categories Without Paired Training Data (CVPR 2016 · similarity 0.90)
- Learning Object Context for Dense Captioning (AAAI 2019 · similarity 0.89)
- Captioning Images With Diverse Objects (CVPR 2017 · similarity 0.89)
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.