Research map / Vision-language and multimodal / Visual question answering
MoVie: Revisiting Modulated Convolutions for Visual Counting and Beyond
Duy Kien Nguyen, Vedanuj Goswami, Xinlei Chen
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 |
|---|---|---|---|
| Replace FiLM modulation with MoVie modulation (ablation) | MoVie counting module · HowMany-QA val | ACC: 57.1 -> 58.4 | Improved |
| Add the learned linear mapping W to modulation (ablation) | FiLM-style modulation · HowMany-QA val | ACC: 57.1 -> 58.4 | Improved |
| Add a residual connection to MoVie modulation (ablation) | MoVie modulation · HowMany-QA val | ACC: 57.9 -> 58.4 | Improved |
| Add beta to MoVie modulation with the residual connection (ablation) | MoVie modulation · HowMany-QA val | ACC: 58.4 -> 58.5 | No clear effect |
| Use multiple modulated bottlenecks instead of one (ablation) | MoVie counting module · HowMany-QA val | ACC: 57.2 -> 58.4 | Improved |
| Use 5 rather than 4 modulated bottlenecks (ablation) | MoVie counting module · HowMany-QA val | RMSE: 2.60 -> 2.63 | Hurt |
| Pad images to a fixed maximum input size (ablation) | MoVie counting module · HowMany-QA val, test size 800 | ACC: 56.2 -> 58.4 | Improved |
| Train at multiple image scales (ablation) | MoVie counting module · HowMany-QA val, test size 800 | ACC: 58.4 -> 58.8 | Improved |
| Integrate MoVie using the three-branch fusion scheme | MCAN-Small · VQA 2.0 val | Number accuracy: 46.71 -> 50.45 | Improved |
| Use naive feature fusion instead of three-branch fusion (ablation) | MCAN-Small + MoVie · VQA 2.0 val | Number accuracy: 46.71 -> 49.36; 50.45 with three-branch fusion | Mixed |
| Add MoVie to MCAN-Large with ResNeXt-101 | MCAN-Large + X-101 · VQA 2.0 test-dev | Number accuracy: 55.68 -> 57.05 | Improved |
| Add self-attention layers to question encoding (ablation) | MoVie question encoder · HowMany-QA val | ACC: 0.7% improvement in accuracy | Improved |
Later papers that compared with this method
1 later accepted paper reports MoVie as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- On Scaling Up a Multilingual Vision and Language Model (CVPR 2024 · better on 1 of 1 shared result)
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
- VLCounter: Text-Aware Visual Representation for Zero-Shot Object Counting (AAAI 2024 · similarity 0.88)
- Unveiling the Visual Counting Bottleneck in Vision-Language Models (ICML 2026 · similarity 0.88)
- CountGD: Multi-Modal Open-World Counting (NeurIPS 2024 · similarity 0.87)
- T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025 · similarity 0.87)
- Understanding Counting Mechanisms in Large Language and Vision-Language Models (CVPR 2026 · similarity 0.87)
- OmniCount: Multi-label Object Counting with Semantic-Geometric Priors (AAAI 2025 · 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.
