AI Paper Atlas

Research map / Vision-language and multimodal / Visual question answering

TAB-VCR: Tags and Attributes based VCR Baselines

Jingxiang Lin, Unnat Jain, Alexander Schwing

NeurIPS 2019 · Vision-language and multimodal › Visual question answering · approach: Visual reasoning diagnostics · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
add Det-BN (train-time batch norm statistics) (ablation)R2C · VCR valQ->AR accuracy: 43.1 -> 43.61Improved
freeze image CNN weights (ablation)R2C + Det-BN · VCR valQ->AR accuracy: 43.61 -> 44.41Improved
use ResNet101 backbone (ablation)R2C++ · VCR valQ->AR accuracy: 44.41 -> 46.42Improved
add attribute features (ablation)R2C++ + ResNet101 · VCR valQ->AR accuracy: 46.42 -> 48.64Improved
use ResNet101 instead of ResNet50 (ablation)Base · VCR valQ->AR accuracy: 46.19 -> 47.51Improved
add attributes (ablation)Base + ResNet101 · VCR valQ->AR accuracy: 47.51 -> 50.08Improved
add new tags for ungrounded nouns (ablation)Base + ResNet101 + Attributes · VCR valQ->AR accuracy: 50.08 -> 50.62Improved
unshare encoder parameters between query and response (ablation)TAB-VCR · VCR valQ->AR accuracy: 50.62 -> 50.35No clear effect
freeze 4th conv block instead of fine-tuning (ablation)Base · VCR valQ->AR accuracy: 44.60 -> 46.45Improved
lower learning rate for fine-tuning conv block (0.5x, 0.25x, 0.125x) (ablation)Base · VCR valQ->AR accuracy: 44.60 -> 44.08Hurt
remove downsample net (ablation)Base (frozen conv block) · VCR valQ->AR accuracy: 46.45 -> 45.57Hurt
more image-text grounding (tags) in query+responseTAB-VCR · VCR valQ->A tags: 2.566 -> 2.719Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy