AI Paper Atlas

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

Learning What Makes a Difference from Counterfactual Examples and Gradient Supervision

Damien Teney, Ehsan Abbasnedjad, Anton van den Hengel

ECCV 2020 · Vision-language and multimodal › Visual question answering · approach: Vision-language model VQA evaluation · 117 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 GS loss on top of counterfactual datastrong VQA baseline · VQA-CP v2 focused test (All)accuracy: 44.2 -> 46.2Improved
add GS loss on top of counterfactual datastrong VQA baseline · VQA-CP v2 test (All)accuracy: 46.0 -> 46.8Improved
add GS loss on top of counterfactual datastrong VQA baseline · VQA-CP v2 val (All)accuracy: 63.3 -> 62.4Hurt
add counterfactual dataweak BUTD baseline (Other only) · VQA-CP v2 test (Other)accuracy: 43.3 -> 45.0Improved
add GS loss on top of counterfactual dataweak BUTD baseline (Other only) · VQA-CP v2 focused test (Other)accuracy: 40.6 -> 38.3Hurt
add edited training imagesCOCO multi-label classifier · COCO edited imagesmAP: 58.1 -> 64.0Improved
add GS loss with counterfactual relationsCOCO classifier with edited data · COCO hard edited imagesmAP: 56.0 -> 57.7Improved
use random relations in GS instead of counterfactual relations (ablation)COCO classifier with GS · COCO original imagesmAP: 71.8 -> 72.9Improved
add edited training datasentiment model · IMDb test editedaccuracy: 55.3 -> 88.7Improved
add GS loss with counterfactual relationssentiment model with edited data · IMDb test editedaccuracy: 88.7 -> 91.2Improved
use counterfactual relations instead of random relations in GS (ablation)sentiment model · IMDb valaccuracy: 50.8 -> 89.8Improved
add edited training dataNLI model · MultiNLI dev (zero-shot)accuracy: 46.0 -> 42.4Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

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

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