Research map / Vision-language and multimodal / Human and embodied modeling
Convolutional Image Captioning
Read the paperCheck 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 |
|---|---|---|---|
| Use a masked-convolution captioning model with attention instead of LSTM | Image captioning · MSCOCO test set | CIDEr: .912 vs .912 | No clear effect |
| Use a masked-convolution captioning model with attention instead of LSTM | Image captioning · MSCOCO test set | BLEU-4: .303 -> .287 | Hurt |
| Use beam search with beam size 3 rather than greedy decoding | CNN+Attn · MSCOCO test set | BLEU-4: .287 -> .316 | Helped |
| Use beam search with beam size 3 rather than greedy decoding | CNN+Attn · MSCOCO test set | CIDEr: .912 -> .952 | Helped |
| Add weight normalization (ablation) | Masked-convolution CNN · MSCOCO val set | CIDEr: .855 -> .881 | Helped |
| Add dropout to the weight-normalized CNN (ablation) | CNN with weight normalization · MSCOCO val set | CIDEr: .881 -> .883 | Helped |
| Add residual connections (ablation) | CNN with weight normalization and dropout · MSCOCO val set | CIDEr: .883 -> .899 | Helped |
| Add attention (ablation) | CNN with weight normalization, dropout, and residual connections · MSCOCO test set | CIDEr: .906 -> .912 | Helped |
| Use ReLU instead of GLU activations (ablation) | CNN convolution layers · MSCOCO | captioning performance | No clear effect |
Closest papers by meaning
- Reference Based LSTM for Image Captioning (AAAI 2017 · similarity 0.91)
- Incorporating Copying Mechanism in Image Captioning for Learning Novel Objects (CVPR 2017 · similarity 0.91)
- Fully Convolutional Video Captioning with Coarse-to-Fine and Inherited Attention (AAAI 2019 · similarity 0.90)
- Partially-Supervised Image Captioning (NeurIPS 2018 · similarity 0.90)
- Adversarial Semantic Alignment for Improved Image Captions (CVPR 2019 · similarity 0.89)
- Unsupervised Image Captioning (CVPR 2019 · 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.