Research map / Vision-language and multimodal / Visual question answering
Revisiting Visual Question Answering Baselines
Allan Jabri, Armand Joulin, Laurens van der Maaten
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 image input to answer+question (ablation) | MLP · Visual7W Telling | accuracy (%): 58.5 -> 67.1 | Improved |
| add question input to answer+image (ablation) | MLP · Visual7W Telling | accuracy (%): 63.8 -> 67.1 | Improved |
| add image and question to answer-only model (ablation) | MLP · Visual7W Telling | accuracy (%): 52.9 -> 67.1 | Improved |
| treat the answer as an input with binary loss instead of softmax over answers (ablation) | MLP · Visual7W Telling | accuracy (%): 52.2 -> 67.1 | Improved |
| treat the answer as an input with binary loss instead of softmax (ablation) | MLP · VQA test2015-dev | accuracy (%): 61.1 -> 64.9 | Improved |
| binary loss instead of softmax (ablation) | Linear · Visual7W Telling | accuracy (%): 42.6 -> 44.7 | Improved |
| use ResNet-101 features instead of AlexNet (ablation) | MLP · Visual7W Telling | accuracy (%): 63.5 -> 67.1 | Improved |
| replace BoW with LSTM text encoder (ablation) | MLP · Visual7W Telling (no image) | accuracy (%): 58.5 -> 52.9 | Hurt |
| replace BoW with LSTM text encoder (ablation) | Bilinear · Visual7W Telling (no image) | accuracy (%): 52.6 -> 54.3 | Improved |
| train on VQA and transfer directly instead of training on Visual7W | MLP (A+Q+I) · Visual7W Telling | accuracy (%): 67.1 -> 53.8 | Hurt |
| add finetuning on Visual7W after VQA transfer | MLP (A+Q+I) · Visual7W Telling | accuracy (%): 67.1 -> 68.5 | Improved |
| use MLP over bilinear model (ablation) | answer-as-input classifier · Visual7W Telling | accuracy (%): 63.6 -> 67.1 | Improved |
Closest papers by meaning
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.95)
- Answer Them All! Toward Universal Visual Question Answering Models (CVPR 2019 · similarity 0.94)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.94)
- VQA: Visual Question Answering (IJCV 2016 · similarity 0.94)
- Answer-Type Prediction for Visual Question Answering (CVPR 2016 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
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.
