Research map / Vision-language and multimodal / Human and embodied modeling
Discriminability Objective for Training Descriptive Captions
Ruotian Luo, Brian Price, Scott Cohen, Gregory Shakhnarovich
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 discriminability loss with λ=1 to ATTN+CIDER (ablation) | ATTN caption generator · COCO test 1k | 4 in 5 human accuracy: 55.95% -> 61.67% | Improved |
| Add discriminability loss with λ=1 to ATTN+CIDER (ablation) | ATTN caption generator · COCO test 1k | BLEU4: 0.3871 -> 0.3971 | Improved |
| Increase discriminability-loss weight from λ=1 to λ=10 (ablation) | ATTN+CIDER+DISC · COCO test 1k | 4 in 5 human accuracy: 61.67% -> 64.63% | Improved |
| Increase discriminability-loss weight from λ=1 to λ=10 (ablation) | ATTN+CIDER+DISC · COCO test 1k | BLEU4: 0.3971 -> 0.3538 | Hurt |
| Add discriminability loss with λ=1 to ATTN+CIDER (ablation) | ATTN caption generator · COCO standard split test | BLEU4: 0.3566 -> 0.3614 | Improved |
| Increase discriminability-loss weight from λ=1 to λ=10 (ablation) | ATTN+CIDER+DISC · COCO standard split test | BLEU4: 0.3614 -> 0.3203 | Hurt |
| Add discriminability loss with λ=1 to ATTN+CIDER (ablation) | ATTN caption generator · COCO standard split test | CIDEr: 1.1345 -> 1.1425 | Improved |
| Add discriminability loss with λ=1 to ATTN+CIDER (ablation) | ATTN caption generator · COCO MSCOCO test server | BLEU-4: 0.349 -> 0.352 | Improved |
| Use discriminability loss with λ=10 instead of λ=1 (ablation) | ATTN+CIDER+DISC · COCO standard split val | CIDEr: 1.1381 -> 1.0402 | Hurt |
| Add discriminability loss with λ=1 to ATTN+CIDER (ablation) | ATTN caption generator · COCO validation | Distinct captions: 2640 -> 3235 | Improved |
Closest papers by meaning
- Show, Tell and Discriminate: Image Captioning by Self-retrieval with Partially Labeled Data (ECCV 2018 · similarity 0.90)
- Generating Diverse and Descriptive Image Captions Using Visual Paraphrases (ICCV 2019 · similarity 0.88)
- Context-Aware Captions From Context-Agnostic Supervision (CVPR 2017 · similarity 0.88)
- Cross-Domain Image Captioning With Discriminative Finetuning (CVPR 2023 · similarity 0.88)
- Improving Image Captioning with Conditional Generative Adversarial Nets (AAAI 2019 · similarity 0.87)
- From Show to Tell: A Survey on Deep Learning-Based Image Captioning (TPAMI 2022 · 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.
