Research map / Image generation / Autoregressive and latent generation
Diverse and Accurate Image Description Using a Variational Auto-Encoder with an Additive Gaussian Encoding Space
Liwei Wang, Alexander Schwing, Svetlana Lazebnik
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 |
|---|---|---|---|
| Use the additive Gaussian prior instead of the fixed Gaussian prior | CVAE captioning · MSCOCO test, oracle | CIDEr: 0.910 -> 1.517 | Improved |
| Use the additive Gaussian prior instead of the Gaussian mixture prior | CVAE captioning · MSCOCO test, oracle | CIDEr: 1.430 -> 1.517 | Improved |
| Use the additive Gaussian prior instead of the Gaussian mixture prior | CVAE captioning · MSCOCO test, diversity | % unique per image: 0.594 -> 0.764 | Improved |
| Increase z samples from 20 to 100 (ablation) | AG-CVAE · MSCOCO test, oracle | CIDEr: 1.308 -> 1.517 | Improved |
| Increase z samples from 20 to 100 (ablation) | AG-CVAE · MSCOCO test, diversity | % unique per image: 0.764 -> 0.550 | Hurt |
| Add beam search with beam size 2 (ablation) | AG-CVAE with 20 z samples · MSCOCO test, oracle | CIDEr: 1.259 -> 1.308 | Improved |
| Add the object vector (ablation) | AG-CVAE with 20 z samples and std 2 · MSCOCO test, oracle | CIDEr: 1.230 -> 1.259 | Improved |
| Use std 2 rather than std 0.1 (ablation) | AG-CVAE without object vector · MSCOCO test, oracle | CIDEr: 1.230 -> 1.230 | No clear effect |
| Replace the additive Gaussian prior with a Gaussian mixture prior (ablation) | CVAE with 100 z samples and object vector · MSCOCO test, oracle | CIDEr: 1.517 -> 1.378 | Hurt |
| Use 100 rather than 20 z samples (ablation) | AG-CVAE · MSCOCO test, diversity | % unique per image: 0.764 -> 0.550 | Hurt |
Closest papers by meaning
- Variational Structured Semantic Inference for Diverse Image Captioning (NeurIPS 2019 · similarity 0.88)
- Patch Matters: Training-free Fine-grained Image Caption Enhancement via Local Perception (CVPR 2025 · similarity 0.86)
- Contrastive Learning for Image Captioning (NeurIPS 2017 · similarity 0.85)
- Beyond a Pre-Trained Object Detector: Cross-Modal Textual and Visual Context for Image Captioning (CVPR 2022 · similarity 0.85)
- Weakly-Supervised Generation and Grounding of Visual Descriptions With Conditional Generative Models (CVPR 2022 · similarity 0.85)
- Modeling Caption Diversity in Contrastive Vision-Language Pretraining (ICML 2024 · similarity 0.84)
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.
